William Vambenepe's blog

Events/alerts/notifications have been a central concept in IT management at least since the first SNMP trap was emitted, and probably even long before that. And yet they are curiously absent from all the Cloud management APIs/protocols. If you think that’s because “THE CLOUD CHANGES EVERYTHING” then you may have to think again. Over the last few days, two of the most experienced practitioners of Cloud computing pointed out that this omission is a real pain in the neck. RightScale’s Thorsten von Eicken was first to request “an event based interface instead of a request-reply based interface”, pointing out that “we run a good number of machines that do nothing but chew up 100% cpu polling EC2 to detect changes”. George Reese seconded and started to sketch a solution. And while these blog posts gave the issue increased visibility recently, it has been a recurring topic on the AWS Forum and other similar discussion boards for quite some time. For example, in this thread going back to 2006, an Amazon employee wrote that “this is a feature we’ve discussed recently and we’re looking at options” (incidentally, I see a post by Thorsten in that old thread). We’re still waiting.

Let’s look at what it would take to define such a feature.

I have some experience with events for IT management, having been involved in the WS-Notification family of specifications and having co-chaired the OASIS technical committee that standardized them. This post is not about foisting WS-Notification on Cloud APIs, but just about surfacing some of the questions that come up when you try to standardize such a mechanism. While the main use cases for WS-Notification came from IT (and Grid) management, it was supposed to be a generic mechanism. A Cloud-centric eventing protocol can be made simpler by focusing on fewer use cases (Cloud scenarios only). In addition, WS-Notification was marred by the complexity-is-a-sign-of-greatness spirit of the time . On this too, a Cloud eventing protocol could improve things by keeping IBM at bay simplicity in mind.

Types of event

When you pull the state of a resource to see if anything changed,  you don’t have to tell the provider what kind of change you are interested in. If, on the other hand, you want the provider to notify you, then they need to know what you care about. You may not want to be notified on every single change in the resource state. How do you describe the changes you care about? Is there an agreed-upon set of states for the resource and you are only notified on state transitions? Can you indicate the minimum severity level for an event to be emitted? Who determines the severity of an event? Or do you get to specify what fields in the resource state you want to watch? What about numeric values for which you may not want to be notified of every change but only when a threshold is crossed? Do you get to specify a query and get notified whenever the query result changes? In WS-Notification some of this is handled by WS-Topics which I still like conceptually (I co-edited it) but is too complex for the task at hand.

Event formats

What format are the events serialized in? How is the even metadata captured (e.g. time stamp of observation, which may not be the same as the time at which the notification message was sent)? If the event payload is a representation of the new state of the resource, does it indicate what field changes (and what the old value was)? How do you keep event payloads consistent with the resource representation in the request/response interactions? If many events occur near the same time, can you group them in one notification message for better scalability?

Subscription creation

Presumably you need a subscription mechanism. Is the subscription set in stone when the resource is created? Or can you come later and subscribe? If subscription is an operation on the resource itself, how do you subscribe for events on something that doesn’t exist yet (e.g. “create a VM and notify me once it’s started”)? Do you get to set subscriptions on a per-resource-basis? Or is this a global setting for all the resources that you own? Can you have two different subscriptions on the same resource (e.g. a “critical events only” subscription that exist throughout the life of the resource, plus a “lots of events please” subscription that you keep for a few hours while troubleshooting)?

Subscription management

Do you get to come back and update/pause/delete a subscription? Do you get to change what filter the subscription carries? Or is it set in stone until the subscription expires? Can you change the delivery endpoint? What if events fail to be delivered? Does the provider cancel your subscription? After how many failures? Does it just pause it for a few hours? Keep trying?

Subscription expiration

Who sets the expiration period? The subscriber? Can the provider set a max duration? Do you get a warning message before the subscription expires? Can you renew a subscription or do you have to create a new one? Do you get a message telling you that it has expired? Where are these subscription-lifecycle messages sent? To the same endpoint as the regular messages? What if your subscription is being killed because your deliver endpoint is down, clearly it makes no sense to send the warning message to that same endpoint. Do you provide a separate “subscription management” endpoint (different from the event delivery endpoint) when you subscribe? Alternatively, does an email message get sent to the registered user who set the subscription?

Delivery reliability

How reliable do you want the notifications to be? Should the emitter retry until they’ve received a confirmation? How long do they keep messages that can’t be delivered? Some may have a very short shelf life while others are still useful weeks later. If you don’t have a reliable mechanism but you really “need to know about a lost server within a minute of it disappearing” (the example Georges gives) then in reality you may still have to poll just to make sure that an event wasn’t lost. If you haven’t received an event in a while, how can you test if the subscription is still working? Should subscriptions send a heartbeat message once a while?

Delivery mechanism

How do you deliver notifications? Do you keep HTTP connections open through tricks similar to how self-updating web pages work (e.g. COMET, long polling and soon WebSockets)? Or do you just provide a listener endpoint to which the notifier tries to connect (which, in the case of public cloud deployments, means you need to have a publicly-addressable listener, but hopefully not on the same Cloud infrastructure). Do you use XMPP? AMQP? Email? Can I have you hold my events and let me come pull them?

Security

Do you need to verify the origin of the events you receive? Or do you assume they may be forged and always initiate a connection to the provider to double-check? And on the other side, what are the security requirements for event delivery? If a user looses some of their privileges, do you have to go and cancel the still-active subscriptions that they created?

Throttling

Is there a maximum event rate? Do you get charged for the events the Cloud provider sends you? How do you make sure that someone doesn’t create a subscription pointing to the wrong endpoint (either erroneously or maliciously, e.g. DoS). Do you send a test message at registration asking the delivery endpoint to acknowledge that they indeed want to receive these notifications?

Conclusion

My goal is not to argue that we cannot have a simple yet good enough notification system or to scare anyone from attempting to define it. It’s just to show that it’s not as simple as it may seem at first blush. But there probably is a sweetspot and people like Thorsten and George are very well qualified to find it.

Oracle just announced that it has purchased Amberpoint. If you have ever been interested in Web services management, then you surely know about Amberpoint. The company has long led the pack of best-of-breed vendors for Web services and SOA Management. My history with them goes back to the old days of the OASIS WSDM technical committee, where their engineers brought to the group a unique level of experience and practical-mindedness.

The official page has more details. In short, Amberpoint is going to reinforce Oracle Enterprise Manager, especially in these areas:

• Business Transaction Management
• SOA Management
• Application Performance Management
• SOA Governance (BTW, Oracle Enterprise Repository 11g was released just over a week ago)

I am looking forward to working with my new colleagues from Amberpoint.

[Preface: a few months ago I shared some thoughts about how REST was (or could) be applied to IT and Cloud management. Part 1 was a comparison of the RESTful aspects of four well-known IaaS Cloud APIs and part 2 was an analysis of how REST applies to configuration management. Both of these entries received well-informed reader comments BTW, so if you read the posts but didn't come back for the comments you really owe it to yourself to do so now. At the time, I jotted down thoughts for subsequent entries in this series, but I never got around to posting them. Since the topic seems to be getting a lot of attention these days (especially in DMTF) I decided to go back to these notes and see if I could extract a few practical recommendations in the form of a wrap-up.]

The findings listed below should be relevant whether your protocol is trying to be truly RESTful, just HTTP-centric or even zen-SOAPy. Many of the issues that arise when creating a protocol that maps well to IT management use cases should transcend these variations and that’s what I try to cover.

Finding #1: Relationships (links) are first-class entities (a.k.a. “hypermedia”)

The clear conclusion of both part 1 and part 2 was that the most relevant part of REST for IT and Cloud management is the use of hypermedia. IT management enjoys a head start on this compared to other domains, because its models are already rich in explicit relationships (e.g. CIM associations), as opposed to other business domains in which relationships are more implicit (to the end user at least). But REST teaches us that just having relationships in your model is not enough. They need to be exposed in a way that maps directly to the protocol, so that following a relationship is an infrastructure-level task, not an application-level task: passing an ID as a parameter for some domain-specific function is not it.

This doesn’t violate the rule to not mix the protocol and the model because the alignment should take place in the metamodel. XML is famously weak in that respect, but that’s where Atom steps in, handling relationships in a generic way. Similarly, support for references is, in addition to its accolade to Schematron, one of the main benefits of SML (extra kudos for apparently dropping the “EPR” reference scheme between submission and standardization, in favor of just the “URI” scheme). Not to mention RDFa and friends. Or HTTP Link headers (explained) for link-challenged types.

Finding #2: Put IDs on steroids

There is little to argue about the value of clearly identifying things of interest and we didn’t wait for the Web to realize this. But it is also one of the most vexing and complex problems in many areas of computing (including IT management). Some of the long-standing questions include:

• Use an opaque ID (some random-looking string a characters) or an ID grounded in “unique” properties of the resource (if you can find any)?
• At what point does a thing stop being the same (typical example: if I replace each hardware component of a server one after the other, at which point is it not the same server anymore? Does it make sense for the IT guys to slap an “asset id” sticker on the plastic box around it?)
• How do you deal with reconciling two resources (with their own IDs) when you realize they represent the same thing?

REST guidelines don’t help with these questions. There often is an assumption, which is true for many web apps, that the application “owns” the resource. My “inbox” only exists as a resource within the mail server application (e.g. Gmail or an Exchange server). Whatever URI GMail assigns for it is the URI for my inbox, period. Things are not as simple when the resources exist outside of any specific application: take a server, for example: the board management controller (or the hypervisor in the case of a VM), the OS management layer and the management agent installed on the machine all have claims to report on the machine (and therefore a need to identify it).

To some extent, Cloud computing simplifies many of these issues by providing controllers that “own” infrastructure resources and can authoritatively identify them. But it really is only pushing the problem to the next level of the stack.

Making the ID a URI doesn’t magically answer these questions. Though it helps in that it lets you leverage reconciliation mechanisms developed around URIs (such as <atom:link rel=”alternate”> or owl:sameAs). What REST does is add another constraint to this ID mechanism: Make the IDs dereferenceable URLs rather than just URIs.

I buy into this. A simple GET on a resource URI doesn’t solve everything but it has so many advantages that it should be attempted in all cases. And make this HTTP GET please (see finding #6).

In this adoption of GET, we just have to deal with small details such as:

• What URL do I use for resources that have more than one agent/controller?
• How close to the resource do I point this URL? If it’s too close to it then it may change as the resource evolves (e.g. network changes) or be affected by the resource performance (e.g. a crashed machine or application that does not respond to its management API). If it’s removed from the resource, then I introduce a scope (e.g. one controller) within which the resource has to remain, which may cause scalability concerns (how many VMs can/should one controller handle, what if I want to migrate a VM across the ocean…).

These are somewhat corner cases (and the more automation and virtualization you get, the fewer possible controllers you have per resource). While they need to be addressed, they don’t come close to negating the value of dereferenceable IDs. In addition, there are plenty of mechanisms to help with the issues above, from links in the representations (obviously) to RDDL-style lightweight directory to a last resort “give Saint Peter a call” mechanism (the original WSRF proposal had a sub-specification called WS-RenewableReferences that would let you ask for a new version of an expired EPR but it was never published — WS-Naming in then-GGF also touched on that with its reference resolvers — showing once again that the base challenges don’t change as fast as technology flavors).

Implicit in this is the fact that URIs are vastly superior to EPRs. The latter were only just a band-aid on a broken system (which may have started back when WSDL 1.1 decided to define “ports” as message aggregators that can have only one URL) and it’s been more debilitating to SOAP than any other interoperability issue. Web services containers internalized this assumption to the point of providing a stunted dispatch mechanism that made it very hard to assign distinct URLs to resources.

Finding #3: If REST told you to jump off a bridge, would you do it?

Adherence to REST is not required to get the benefits I describe in this series. There is a lot to be inspired by in REST, but it shouldn’t be a religion. Sure, if you squint hard enough (and poke it here and there) you can call your interface RESTful, but why bother with the contortions if some parts are not so. As long as they don’t detract from the value of REST in the other parts. As in all conversions, the most fervent adepts of RPC will likely be tempted to become its most violent denunciators once they’re born again. This is a tired scenario that we don’t need to repeat. Don’t think of it as a conversion but as a new perspective.

Look at the “RESTful with many parameters?” comment thread on Stefan Tilkov’s excellent InfoQ introduction to REST. It starts with some shared distaste for parameter-laden URIs and a search for a more RESTful approach. This gets suggested:

You could do a post on some URI like ./query/product_dep which would create a query resource. Now you “add” products to the query either by sending a product uri list with the initial post or by calling post on ./query/product_dep/{id}. With every post to the query resource the get on the query resource would change.

Yeah, you could. But how about an RPC-like query operation rather than having yet another resource lifecycle to manage just for the sake of being REST-compliant? And BTW, how do you think any sane consumer of your API is going to handle this? You guessed it, by packaging the POST/POST/GET/DELETE in one convenient client-side library function called “query”. As much as I criticize RPC-centric toolkits (see finding #5 below), it would be justified in this case.

Either you understand why/how REST principles benefit you or you don’t. If you do, then use this understanding to interpret the REST principles to best fit your needs. If you don’t, then no amount of CONTENT-TYPE-pixie-dust-spreading, GET-PUT-POST-DELETE-golden-rule-following and HATEOAS-magical-incantation-reciting will help you. That’s the whole point, for me at least, of this tree-part investigation. Stefan says essential the same, but in a converse way, in his article: “there are often reasons why one would violate a REST constraint, simply because every constraint induces some trade-off that might not be acceptable in a particular situation. But often, REST constraints are violated due to a simple lack of understanding of their benefits.” He says “understand why you violate” and I say “understand why you obey”. It is essentially the same (if you’re into stereotypes you can attribute the difference to his Germanic heritage and my Gallic blood).

Even worse than bending your interface to appear RESTful, don’t cherry-pick your use cases to only keep those that you feel you can properly address via REST, leaving the others aside. Conversely, don’t add requirements just because REST makes them easy to support (interesting how quickly “why do you force me to manage the lifecycle of yet another resource just to run a query” turns into “isn’t this great, you can share queries among users and you can handle long-running queries, I am sure we need this”).

This is not to say that you should not create a fully RESTful system. Just that you don’t necessarily have to and you can still get many benefits as long as you open your eyes to the cost/benefits trade-off involved.

Finding #4: Learn humility from REST

Beyond the technology, there is a vibe behind REST design. You can copy the technology and still miss it. I described it in 2005 as Humble Architecture, and applied to SOA at the time. But it describes REST just as well:

More practically, this means that the key things to keep in mind when creating a service, is that you are not at the center of the universe, that you don’t know who is going to consume your service, that you don’t know what they are going to do with it, that you are not necessarily the one who can make the best use of the information you have access to and that you should be willing to share it with others openly…

The SOA Manifesto recently called this “intrinsic interoperability”.

In IT management terms, it means that you can RESTify your CMDB and your event console and your asset management software and your automation engine all you want, if you see your code as the ultimate consumer and the one that knows best, as the UI that users have to go through, the “ultimate source of truth” and the “manager of managers” then it doesn’t matter how well you use HTTP.

Finding #5: Beware of tools bearing gifts

To a large extent, the great thing about REST is how few tools there are to take it away from you. So you’re pretty much forced to understand what is going on in your contract as opposed to being kept ignorant by a wsdl2java type of toolkit. Sure, Java (and .NET) have improved in that regard, but really the cultural damage is done and the expectations have been set. Contrast this to “the ‘router’ is just a big case statement over URI-matching regexps”, from Tim Bray’s post on the Sun Cloud API, one of my main inspirations for this investigation.

REST is not inherently immune to the tool-controlling-the-hand syndrome. It’s just a matter of time until such tools try to make REST “accessible” to the “normal” developer (who can supposedly prevent thread deadlocks but not parse XML). Joe Gregorio warns about this in the context of WADL (to summarize: WADL brings XSD which leads to code generation). Keep this in mind next time someone states that REST is more “loosely coupled” than SOAP. It’s how you use it that matters.

Finding #6: Use screws, not glue, so we can peer inside and then close the lid again

The “view source” option is how I and many others learned HTML. It unfortunately created a generation of HTML monsters who never went past version 3.2 (the marbled background makes me feel young again). But it also fueled the explosion of the Web. On-the-wire inspection through soapUI is what allowed me to perform this investigation and report on it (WMI has allowed this for years, but WS-Management is what made it accessible and usable for anyone on any platform). This was, of course, in the context of SOAP which is also inspectable. Still, in that respect nothing beats plain HTTP which is why I recommend HTTP GET in finding #2 (make IDs dereferenceable) even though I don’t expect that the one-page-per-resource view is going to be the only way to access it in the finished product.

These (HTML source, on-the-wire XML and resource-description pages) rarely hit the human eye and yet their presence enables the development of the more commonly used views. Making it as easy as possible to see what is going on under the covers helps with learning, with debugging, with extending and with innovating. In the same way that 99% of web users don’t look at the HTML source (and 99.99% of them don’t see the HTTP requests) but the Web would not be what it is to them if this inspectability wasn’t been there to fuel its development.

Along the same line, make as few assumptions as possible about the consumers in your interfaces. Which, in practice, often means document what goes on the wire. WSDL/WADL can be used as a format, but they are at most one small component. Human-readable semantics are much more important.

Finding #7: Nothing is free

Part of what was so attractive about SOAP is everything you were going to get “for free” by using it. Message-level security (for all these use cases where your messages starts over HTTP, then hops onto a train, then get delivered by a carrier pigeon). Reliable messaging. Transactionality. Intermediaries (they were going to be a big deal in SOAP, as you can see in vestigial form today in the Nodes/Roles left in the spec – also, do you remember WS-Routing? I do.)

And it’s true that by now there is a body of specifications that support this as composable SOAP headers. But the lack of usage of these features contrasts with how often they were bandied in the early days of SOAP.

Well, I am detecting some of the same in the REST camp. How often have you heard about how REST enables caching? Or about how content types allows an ISP to compress images on the fly to speed up delivery over dial-up? Like in the SOAP case, these are real features and sometimes useful. It doesn’t mean that they are valuable to you. And if they are not, then don’t let them be used as justifications. Especially since they are not free. If caching doesn’t help me (because of low volume, because security considerations prevent a shared cache, etc) then its presence actually adds a cost to me, since I now have to worry whether something is cached or not and deal with ETags. Or I have to consistently remember to request the cache to be bypassed.

Finding #8: Starting by sweeping you front door.

Before you agonize about how RESTful your back-end management protocol is, how about you make sure that your management application (the user front-end) is a decent Web application? One with cool URIs , where the back button works, where bookmarks work, where the data is not hidden in some over-encompassing Flash/Silverlight thingy. Just saying.

***

Now for some questions still unanswered.

Question #1: Is this a flee market?

I am highly dubious of content negotiation and yet I can see many advantages to it. Mostly along the lines of finding #6: make it easy for people to look under the hood and get hold of the data. If you let them specify how they want to see the data, it’s obviously easier.

But there is no free lunch. Even if your infrastructure takes care of generating these different views for you (“no coding, just check the box”), you are expanding the surface of your contract. This means more documentation, more testing, more interoperability problems and more friction when time comes to modify the interface.

I don’t have enough experience with format negotiation to define the sweetspot of this practice. Is it one XML representation and one HTML, period (everything else get produced by the client by transforming the XML)? But is the XML Atom-wrapped or not? What about RDF? What about JSON? Not to forget that SOAP wrapper, how hard can it be to add. But soon enough we are in legacy hell.

Question #2: Mime-types?

The second part of Joe Gregorio’s WADL entry is all about Mime types and I have a harder time following him there. For one thing, I am a bit puzzled by the different directions in which Mime types go at the same time. For example, we have image formats (e.g. “image/png”), packaging/compression formats (e.g. “application/zip”) and application formats (e.g. “application/vnd.oasis.opendocument.text” or “application/msword”). But what if I have a zip full of PNG images? And aren’t modern word processing formats basically a zip of XML files? If I don’t have the appropriate viewer, maybe I’d like them to be at least recognized as ZIP files. I don’t see support for such composition and taxonomy in these types.

And even within one type, things seem a bit messy in practice. Looking at the registered applications in the “options” menu of my Firefox browser, I see plenty of duplication:

• application/zip vs. application/x-zip-compressed
• application/ms-powerpoint vs. application/vnd.ms-powerpoint
• application/sdp vs. application/x-sdp
• audio/mpeg vs. audio/x-mpeg
• video/x-ms-asf vs. video/x-ms-asf-plugin

I also wonder at what level of depth I want to take my Mime types. Sure I can use Atom as a package but if the items I am passing around happen to be CIM classes (serialized to XML), doesn’t it make sense to advertise this? And within these classes, can I let you know which domain (e.g. which namespace) my resources are in (virtual machines versus support tickets)?

These questions may simply be a reflection of my lack of maturity in the fine art of using Mime types as part of protocol design. My experience with them is more of the “find the type that works through trial and error and then leave it alone” kind.

[Side note: the first time I had to pay attention to Mime types was back in 1995/1996, playing with non-parsed headers and the multipart/x-mixed-replace type to bring some dynamism to web pages (that was before JavaScript or even animated GIFs). The site is still up, but the admins have messed up the Apache config so that the CGIs aren't executed anymore but return the Python code. So, here are some early Python experiments from yours truly: this script was a "pushed" countdown and this one was a "pushed" image animation. Cool stuff at the time, though not in a "get a date" kind of way.]

On the other hand, I very much agree with Joe’s point that “less is more”, i.e. that by not dictating how the semantics of a Mime type are defined the system forces you to think about the proper way to define them (e.g. an English-language RFC). As opposed to WSDL/XSD which gives the impression that once your XML validator turns green you’re done describing your interface. These syntactic validations are a complement at best, and usually not a very useful one (see “fat-bottomed specs”).

In comments on previous posts, Stu Charlton also emphasizes the value that Mime types bring. “Hypermedia advocates exposing a variety of links for such state-transitions, along with potentially unique media types to describe interfaces to those transitions.” I get the hypermedia concept, the HATEOAS approach and its very practical benefits. But I am still dubious about the role of Mime types in achieving them and I am not the only one with such qualms. I have too much respect for Joe and Stu to dismiss it entirely, but until I get an example that makes it “click” in practice for me I won’t sweat about Mime types too much.

Question #3: Riding the Zeitgeist?

That’s a practical question rather than a technical one, but as a protocol creator/promoter you are going to have to decide whether you market it as “RESTful”. If I have learned one thing in my past involvement with standards it is that marketing/positioning/impressions matter for standards as much as for products. To a large extent, for Clouds, Linked Data is a more appropriate label. But that provides little marketing/credibility humph with CIOs compared to REST (and less buzzword-compliance for the tech press). So maybe you want to write your spec based on Linked Data and then market it with a REST ribbon (the two are very compatible anyway). Just keep in mind that REST is the obvious choice for protocols in 2009 in the same way that SOAP was a few years ago.

Of course this is not an issue if you specification is truly RESTful. But none of the current Cloud “RESTful” APIs is, and I don’t expect this to change. At least if you go by Roy Fielding’s definition (or Paul’s handy summary):

A REST API must not define fixed resource names or hierarchies (an obvious coupling of client and server). Servers must have the freedom to control their own namespace. Instead, allow servers to instruct clients on how to construct appropriate URIs, such as is done in HTML forms and URI templates, by defining those instructions within media types and link relations. [Failure here implies that clients are assuming a resource structure due to out-of band information, such as a domain-specific standard, which is the data-oriented equivalent to RPC's functional coupling].

And (in a comment) Mark Baker adds:

I’ve reviewed lots of “REST APIs”, many of them privately for clients, and a common theme I’ve noticed is that most folks coming from a CORBA/DCE/DCOM/WS-* background, despite all the REST knowledge I’ve implanted into their heads, still cannot get away from the need to “specify the interface”. Sometimes this manifests itself through predefined relationships between resources, specifying URI structure, or listing the possible response codes received from different resources in response to the standard 4 methods (usually a combination of all those). I expect it’s just habit. But a second round of harping on the uniform interface – that every service has the same interface and so any service-specific interface specification only serves to increase coupling – sets them straight.

So the question of whether you want to market yourself as RESTful (rather than just as “inspired by the proper use of HTTP illustrated by REST”) is relevant, if only because you may find the father of REST throwing (POSTing?) tomatoes at you. There is always a risk in wearing clothes that look good but don’t quite fit you. The worst time for your pants to fall off is when you suddenly have to start running.

For more on this, refer to Ted Neward’s excellent Roy decoder ring where he not only explains what Roy means but more importantly clarifies that “if you’re not doing REST, it doesn’t mean that your API sucks” (to which I’d add that it is actually more likely to suck if you try to ape REST than if you allow yourself to be loosely inspired by it).

***

Wrapping up the wrap-up

There is one key topic that I had originally included in this wrap-up but decided to remove: extensibility. Mark Hapner brings it up in a comment on a previous post:

It is interesting to note that HTML does not provide namespaces but this hasn’t limited its capabilities. The reason is that links are a very effective mechanism for composing resources. Rather than composition via complicated ‘embedding’ mechanisms such as namespaces, the web composes resources via links. If HTML hadn’t provided open-ended, embeddable links there would be no web.

I am the kind of guy who would have namespace-qualified his children when naming them (had my wife not stepped in) so I don’t necessarily see “extension via links” as a negation of the need for namespaces (best example: RDF). The whole topic of embedding versus linking is a great one but this post doesn’t need another thousand words and the “REST in practice” umbrella is not necessarily the best one for this discussion. So I hereby conclude my “REST in practice for IT and Cloud management” series, with the intent to eventually start a “Linked Data in practice for IT and Cloud management” series in which extensibility will be properly handled. And we can also talk about querying (conspicuously absent from Cloud APIs, unless CMDBf is now a Cloud API) and versioning. As a teaser for the application of Linked Data to IT/Cloud, I will leave you with what Vint Cerf has to say.

[UPDATED 2010/1/27: I still haven't written the promised "Linked Data in practice for IT and Cloud management" post, but this explanation of the usage of Linked Data for data.gov.uk pretty much says it all. I may still write a post describing how what Jeni says about government data applies to Cloud management APIs, but it's almost too obvious to bother. Actually, there may be reasons why Cloud management benefits even more from Linked Data than UK government data, so it may still be worth a post. At some point. When I convince myself that it may influence things rather than be background noise.]

PaaS can most dramatically improve the IT experience in four areas:

• Hosting/operations efficiency
• Application-centric management
• Development productivity
• Security

To do so, there are technical characteristics that PaaS frameworks should eventually exhibit. These are not technical characteristics of a given PaaS container, they are shared characteristics that go across all container types, no matter what the operational capabilities of the containers are.

Here is a rough and unorganized list of the desirable characteristics (meta-capabilities) of PaaS Cloud containers:

• An application component model that supports deployment/configuration across all PaaS container types.
• Explicit interactions/invocations between application components (resilient connections between component: infrastructure-level retry/reroute)
• Uniform and consistent request tracking across all components. Ability to intercept component-to-component communication.
• Short-term (or externally persisted) state so that all instances can be quickly redirected out of any one node.
• Subset of platform management interface exposed to consumer, along with out of the box application management. Application metrics consolidated at application level rather than node level.
• Consistent, model-based application management interface across all container types. Hooks for component code to provide its manageability in the same framework.
• Minimal footprint of any container node for limited patching requirements.
• Assistance for debugging platform-hosted code (see this entry).
• No encroachment of container technology on application contract (e.g. no forced URL structure).
• Application uniformly scalable to the limit of the underlying hardware (no imposed partitioning).
• Shared authentication / authorization / auditing across containers.
• Minimum contract/interface exposed by each container.
• Governance of application services, aligned (in model/protocols) with the container management interfaces.
• [UPDATE: need to add metering+billing as William Louth pointed out in a comment]

This applies across the board to public, private and hybrid PaaS. The distinctions between these delivery models are real but at a different level. The important thing is that the PaaS administrator is different from the application administrator in all cases. On the other hand, most of these technical characteristics are not achievable for lower-level Cloud resources (like virtual hosts and low-level storage) which is why the IaaS form of Cloud leaves the Cloud promise only partially fulfilled.

Andi Mann recently wrote an interesting post about virtual appliances . He uses the domain name pleasediscuss.com for his blog so I figured I’d do just that. More specifically, I have three comments on his article.

Opaque or transparent appliance

Andi’s concerns about the security and management problems posed by virtual appliances are real, but he seems to assume that the content of the appliance is necessarily opaque to the customer and under the responsibility of the appliance provider. Why can’t a virtual appliance be transparent in the sense that the customer is able to efficiently manage at least some aspects of the software installed on it? “You can’t put agents on most virtual appliances, they don’t come with WMI, and most have only a GUI for management” says Andi. Why can’t an appliance come with an agent (especially in these days of consolidation where many vendors provide many layers of the stack – hypervisor / OS / application container / application / management tools – including their agent)? Why can’t it implement a standard management API (most servers nowadays implement WBEM, WS-Management and/or IPMI pre-boot – on the motherboard – which is a lot more challenging to do than supporting a similar protocol in a virtual appliance). Andi is really criticizing the current offering more than the virtual appliance model per se and in this I can join him.

Let me put it differently, since this is probably just a question of definition: what would Andi call a virtual appliance that does expose management APIs for its infrastructure (e.g. WS-Management for the OS, JMX for the java stack) or that comes with an agent (HP, IBM, BMC, Oracle…) installed on it?

Such an appliance (let’s call it a “transparent virtual appliance” for now) doesn’t provide all the commonly claimed benefits of an appliance (zero config/admin) but as Andi points out these benefits come with major intrinsic drawbacks. A transparent virtual appliance still drastically simplifies installation (especially useful for test/dev/demo/POC). It doesn’t entirely free you of monitoring and configuration but at least it provides you with a very consistent and controlled starting point, manageable from the start (no need to subsequently install an agent). In addition, it can be made “just enough” (just enough OS, just enough app server…) to require a lot less maintenance than an application stack that you assemble yourself out of generic parts. We’ll always have trade offs between how optimized/customized it is versus how uniform your overall environment can be, but I don’t see the use of an appliance as a delivery mechanism as necessarily cornering you into a completely opaque situation, from a management perspective.

Those who attended Oracle Open World a few weeks ago were treated to an example of such an appliance, if they attended any of the sessions that covered Oracle’s Appliance Builder (the main one was, I believe, Virtualizing Oracle Fusion Middleware in the Modern Data Center, in case you have access to the Open World On Demand replay and slides). I believe it’s probably the same content that @jayfry3 was shown when he tweeted about “Oracle is demoing their private cloud self-service app”. These appliances are not at all opaque from a management perspective. To the contrary, they are highly manageable, coming with an Enterprise Manager agent installed that can manage everything in the appliance (and when that “everything” doesn’t include the OS, it’s because there isn’t one thanks to JRockit Virtual Edition, making things slimmer, faster, safer and more manageable). And of course the OVM-based environment in which you deploy these appliances is also managed by Enterprise Manager. OK, my point here wasn’t to go into marketing mode, but this is cool stuff and an example of what virtual appliances should be. BTW, this was also demonstrated during Hasan Rizvi’s keynote at OpenWorld, including the management of these systems through Enterprise Manager.

In the long run it’s irrelevant

As with all things computer-related, the issue is going to get blurrier and then irrelevant . The great thing about software is that there is no solid line. In this case, we will eventually get more customized appliances (via appliance builders or model-driven appliance generation) blurring the line between installed software and appliance-based software.

Waiting for PaaS

Towards the end of his post, Andi paints an optimistic vision of the future: “I also think that virtual appliances have a bright future – but in some ways I continue to see them as a beta version of what could (or should) come next.  By adding in capabilities for responsible and accountable management, they could form the basis of more fully-functional virtual service management containers. These in turn could form the basis of elastic, mobile, network-deployed, responsible cloud appliances that deliver complete end-to-end service management without regard to physical location or domain of control.”

I mostly agree with this vision, though when I describe it it is in the guise of a PaaS platform. Where your appliance (which today goes from the OS all the way to the app) has shrunk to an application template that you deploy in the PaaS environment (rather than in a hypervisor). If/when the underlying PaaS environment has reached the right level of management automation you get all the benefits of an appliance while maintaining the consistency of your environment and its adherence to your management policies (because the environment is the PaaS platform and its management is driven from your policies).

[As is often the case, this started as a comment (on Andi's blog) and quickly outgrew that environment, leading to this new post. Plus, Andi's blog is brand new and seems to be well worth spreading the word about (Andi himself is under-marketing it).]

The potential user impact of changes (e.g. patches or config changes) made on the Cloud infrastructure (by the Cloud provider) is a sore point in the Cloud value proposition (see Hoff’s take for example). You have no control over patching/config actions taken by the provider, any of which could potentially affect you. In a traditional data center, you can test the various changes on specific applications; you don’t have to apply them at the same time on all servers; and you can even decide to skip some infrastructure patches not relevant to your application (“if it aint’ broken…”). Not so in a Cloud environment, where you may not even know about a change until after the fact. And you have no control over the timing and the roll-out of the patch, so that some of your instances may be running on patched nodes and others may not (good luck with troubleshooting that).

Unfortunately, this is even worse for PaaS than IaaS. Simply because you seat on a lot more infrastructure that is opaque to you. In a IaaS environment, the only thing that can change is the hardware (rarely a cause of problem) and the hypervisor (or equivalent Cloud OS). In a PaaS environment, it’s all that plus whatever flavor of OS and application container is used. Depending on how streamlined this all is (just enough OS/AS versus a traditional deployment), that’s potentially a lot of code and configuration. Troubleshooting is also somewhat easier in a IaaS setup because the error logs are localized (or localizable) to a specific instance. Not necessarily so with PaaS (and even if you could localize the error, you couldn’t guarantee that your troubleshooting test runs on the same node anyway).

In a way, PaaS is squeezed between IaaS and SaaS on this. IaaS gets away with a manageable problem because the opaque infrastructure is not too thick. For SaaS it’s manageable too because the consumer is typically either a human (who is a lot more resilient to change) or a very simple and well-understood interface (e.g. IMAP or some Web services). Contrast this with PaaS where the contract is that of an application container (e.g. JEE, RoR, Django).There are all kinds of subtle behaviors (e.g, timing/ordering issues) that are not part of the contract and can surface after a patch: for example, a bug in the application that was never found because before the patch things always happened in a certain order that the application implicitly – and erroneously – relied on. That’s exactly why you always test your key applications today even if the OS/AS patch should, in theory, not change anything for the application. And it’s not just patches that can do that. For example, network upgrades can introduce timing changes that surface new issues in the application.

And it goes both ways. Just like you can be hurt by the Cloud provider patching things, you can be hurt by them not patching things. What if there is an obscure bug in their infrastructure that only affects your application. First you have to convince them to troubleshoot with you. Then you have to convince them to produce (or get their software vendor to produce) and deploy a patch.

So what are the solutions? Is PaaS doomed to never go beyond hobbyists? Of course not. The possible solutions are:

• Write a bug-free and high-performance PaaS infrastructure from the start, one that never needs to be changed in any way. How hard could it be? ;-)
• More realistically, narrowly define container types to reduce both the contract and the size of the underlying implementation of each instance. For example, rather than deploying a full JEE+SOA container componentize the application so that each component can deploy in a small container (e.g. a servlet engine, a process management engine, a rule engine, etc). As a result, the interface exposed by each container type can be more easily and fully tested. And because each instance is slimmer, it requires fewer patches over time.
• PaaS providers may give their users some amount of visibility and control over this. For example, by announcing upgrades ahead of time, providing updated nodes to test on early and allowing users to specify “freeze” periods where nothing changes (unless an urgent security patch is needed, presumably). Time for a Cloud “refresh” in ITIL/ITSM-land?
• The PaaS providers may also be able to facilitate debugging of infrastructure-related problem. For example by stamping the logs with a version ID for the infrastructure on the node that generated the log entry. And the ability to request that a test runs on a node with the same version. Keeping in mind that in a SOA / Composite world, the root cause of a problem found on one node may be a configuration change on a different node…

Some closing notes:

• Another incarnation of this problem is likely to show up in the form of PaaS certification. We should not assume that just because you use a PaaS you are the developer of the application. Why can’t I license an ISV app that runs on GAE? But then, what does the ISV certify against? A given PaaS provider, e.g. Google? A given version of the PaaS infrastructure (if there is such a thing… Google advertises versions of the GAE SDK, but not of the actual GAE runtime)? Or maybe a given PaaS software stack, e.g. the Oracle/Microsoft/IBM/VMWare/JBoss/etc, meaning that any Cloud provider who uses this software stack is certified?
• I have only discussed here changes to the underlying platform that do not change the contract (or at least only introduce backward-compatible changes, i.e. add APIs but don’t remove any). The matter of non-compatible platform updates (and version coexistence) is also a whole other ball of wax, one that comes with echoes of SOA governance discussions (because in PaaS we are talking about pure software contracts, not hardware or hardware-like contracts). Another area in which PaaS has larger challenges than IaaS.
• Finally, for an illustration of how a highly focused and specialized container cuts down on the need for config changes, look at this photo from earlier today during the presentation of JRockit Virtual Edition at Oracle Open World. This slide shows (in font size 3, don’t worry you’re not supposed to be able to read), the list of configuration files present on a normal Linux instance, versus a stripped-down (“JeOS”) Linux, versus JRockit VE.

By the way, JRockit VE is very interesting and the environment today is much more favorable than when BEA first did it, but that’s a topic for another post.

[UPDATED 2009/10/22: For more on this (in an EC2-centric context) see section 4 ("service problem resolution") of this IBM paper. It ends with "another possible direction is to develop new mechanisms or APIs to enable cloud users to directly and automatically query and correlate application level events with lower level hardware information to better identify the root cause of the problem".]

Remember 2006? Things were starting to fall into place for IT management integration and automation:

• SDD was already on its way to cleanly describe/package/manage the lifecycle of simple and composite applications alike,
• the first version of SML came out to capture all the relevant constraints of complex and composite systems and open the door to “desired-state management”,
• the CMDBf effort was started to seamlessly integrate all sources of configuration and provide a bird-eye view of your entire IT infrastructure, and
• the WSDM/WS-Management convergence/reconciliation was announced and promised to free management consoles from supporting many resource discovery, collection and control mechanisms and from having platform/library dependencies between the manager and its targets.

It looked like we were a year or two from standardization on all these and another year or two from shipping implementations. Things were looking good.

Good news: the schedule was respected. SDD, SML and CMDBf are now all standards (at OASIS, W3C and DMTF respectively). And today the Eclipse COSMOS project announced the release of COSMOS 1.1 which implements them all. The WSDM/WS-Management convergence is the only one that didn’t quite go according to the plan but it is about to come out as a standard too (in a pared-down form).

Bad news: nobody cares. We’ve moved on to “private clouds”.

Having been involved with these specifications in various degrees (a little bit on SDD, a fair amount on SML and a lot on CMDBf and WSDM/WS-Management) I am not as detached as my sarcastic tone may suggest. But as they say in action movies, “don’t let sentiments get in the way of the mission”.

There is still a chance to reuse parts of this stack (e.g. the CMDBf query language) and there are lessons to learn from our errors. The over-promising, the technical misjudgments, the political bickering, the lack of concrete customer validation, etc. To some extent this work was also victim of collateral damages from the excesses of WS-* (I am looking at you WS-Addressing). We also failed to notice the rise of the hypervisor in our peripheral vision.

I tried to capture some important lessons in this post-mortem. For the edification of the cloud generation. I also see a pendulum in action. Where we over-engineered I now see some under-engineering (overly granular interaction models, overemphasis on the virtual machine as the unit of everything, simplistic constraint models, underestimation of config/patching issues…). Things will come around and may eventually look familiar (suggested exercise: compare PubSubHubBub with WS-Notification).

As long as each iteration gets us closer to the goal things are good.

See you in 2012. Same place, same day, same time.

PHP developers with Cloud aspirations rejoice! Zend has announced a PHP toolkit (called the Simple Cloud API project) to abstract and access application-level Cloud services. This is not just YACA (yet another Cloud API), as there are interesting differences between this and all the other Cloud toolkits out there.

First it’s PHP, which was not covered by the existing toolkits. Considering how many web applications are written in PHP (including the one that serves this very blog) this may seem strange, until you realize that most Cloud toolkits out there are focused on provisioning/managing low-level compute resources of the IaaS kind. Something that is far out of PHP’s sweetspot and much more practically handled with Java, Python, Ruby or some .NET language accessible via PowerShell.

Which takes us to the second, and arguably most interesting, characteristic of this toolkit: it is focused on application-level Cloud services (files, documents and queues for now) rather than infrastructure-level. In other word, it’s the first (to my knowledge) PaaS toolkit.

I also notice that Zend has gotten endorsements from IBM, Microsoft, Nirvanix, Rackspace and GoGrid. The first two especially seem to have impressed InfoWorld. Let’s keep in mind that at this point all we are talking about are canned quotes in a press release. Which rank only above politician campaign promises as predictor of behavior. In any case that can’t be the full extent of IBM and Microsoft’s response to the VMWare/Cisco push on IaaS standards. But it may suggest that their response will move the battlefield to include PaaS, which would be a smart move.

Now for a few more acerbic comments:

• It has “simple” in its name, like SOAP (as Pete Lacey famously lampooned). In the long term this tends to negatively correlate with simplicity, just like the presence of “democratic” in the official name of a country does not bode well for actual democracy.
• Please, don’t shorten “Simple Cloud API” to SCA which is already claimed in a (potentially) closely related field.
• Reuven Cohen is technically correct to see it as “a way to create other higher level programmatic API interfaces such as REST or SOAP using an easy, yet portable PHP programming environment”. But pay attention to how many turtles are on this pile: the native provider API, the adapter to the “simple cloud API”, the SOAP or REST remote API and the consuming application’s native API. How much real isolation are you getting when you build your house on such a wobbly foundation

[UPDATE: Comments from someone in the know:  a programmer working on adding Azure support for this Simple Cloud API project.]

Cloud development toolkits like Libcloud (for Python) and jcloud (for Java) have been around for some time, but over the last two months they have been joined by several other open source contenders. They all claim to abstract the on-the-wire Cloud management protocols sufficiently to let you access different Clouds via the same code; while at the same time providing objects in your programming language of choice and saving you the trouble of dealing with on-the-wire messages. By focusing on interoperability, they slot themselves below the larger role of a “Cloud broker” (which also deals with tasks like transfer and choice). Here is the list, starting with the more recent contenders:

• CloudLoop (here is the announcement) – focused on storage
• OpenNebula (here is the announcement)
• Dasein (here is the announcement)
• jclouds
• Libcloud

DeltaCloud shares the same goal of translating between different Cloud management protocols but they present their own interface as yet another Cloud REST API/protocol rather than a language-specific toolkit. More along the lines of what UCI is trying to do (not sure what’s up with that project, I recorded my skepticism earlier and am still waiting to be pleasantly surprised).

Of course there are also programming toolkits that are specific to one Cloud provider. They are language-specific wrappers around one Cloud management protocol. AWS protocols (EC2, S3, etc…) represent the most common case, for example amazon-ec2 (a Ruby Gem), Power-EC2Dream (in C# which gives it the tantalizing advantage of being invokable via PowerShell) and typica (for Java). For Clouds beyond AWS, check out the various RightScale Ruby Gems.

The main point of this entry was to list the cross-Cloud development toolkits in the bullet list above. But if you’re in the mood for some pontification you can keep reading.

For some reason, what used to be called “protocols” is often called “APIs” in Cloud settings. Witness the Sun Cloud “API” or the vCloud “API” which only define XML formats for on-the-wire messages. I have never heard of CIM/XML over HTTP, WSDM or WS-Management being referred as APIs though they occupy a very similar place. They are usually considered “protocols”.

It’s a just question of definition whether an on-the-wire protocol (rather than a language-specific set of objects/methods) qualifies as an “Application Programming Interface”. It’s not an “interface” in the Java sense of the term. But I can “program” against it so it could go either way. On this blog I have gone along with the “API” term because that seemed widely used, though in verbal conversations I have tended to stick to “protocol”. One problem with “API” is that it pushes you towards mixing the “what” and the “how” and not respecting the protocol/model dichotomy.

Where is becomes relevant is when you start to see language-specific APIs for Cloud control pop-up as listed above. You now have two classes of things called “API” and it gets a bit confusing. Is it time to bring back the “protocol” term for on-the-wire definitions?

As a developer, whether you’re better off eating your Cloud noodles using chopsticks (on-the-wire protocol definitions) or a fork (language-specific APIs) is an important decision that will stay with you and may come back to bit you (e.g. when the interfaces are versioned). There is a place for both of course, but if we are to learn anything from WS-* it’s that we went way too far in the “give me a java stub” direction. Which doesn’t mean there is no room for them, but be careful how far from the wire semantics you get. It become even trickier when your stub tries not jsut to bridge between XML and Java but also to smooth out the differences between several on-the-wire protocols, as the toolkits above do. The hope, of course, is that there will eventually be enough standardization of on-the-wire protocols to make this a moot point.

What happened to the separation between the model and the protocol in management APIs? For all the arguments we had in the design of WSDM and WS-Management, this was one fundamental concept that took little discussion before everyone agreed: that the protocol (the interaction model and the on-the-wire shape of the messages used) should be defined in a way that is agnostic to the type of resource being managed (computers, elevators or toasters — the perennial silly example). To this end, WSDM took pains to release MUWS (Management Using Web Services) and MOWS (Management Of Web Services) as two different specifications.

Contrast that to the different Cloud APIs (there is a new one released every other day). If they have one thing in common it is that they happily ignore this principle and tackle protocol concerns alongside the resource model. Here are my guesses as to why that is:

1) It’s a land grad

The goal is not to produce the best long-term API, it’s to be out early, to stake your claim and to gain leverage, so that you can steer the final standard close to your implementation. Editorial niceties like properly factoring the specification are not major concerns, there will be plenty of time for this during the standardization process. In fact, leaving such improvements for the standardization phase is a nice way to make it look like the group is not just rubberstamping, while not changing much that actually impacts your implementation. The good old “give them something insignificant to argue about” trick. It works BTW.

As an example of how rushed some of these submissions can be, did you notice that what VMWare submitted to DMTF this week is the vCloud API Specification v0.8 (a 7-page document that is simply a list of operations), not the accompanying vCloud API programming guide v0.8 which is ten times longer and is the real specification, the place where the operation semantics, payload formats and protocol considerations are actually described and without which the previous document cannot possibly be implemented. Presumably the VMWare team was pressed to release on time for a VMWorld announcement and they came up with this to be able to submit without finishing all the needed editorial work. I assume this will follow soon and in the meantime the DMTF members will retrieve the programming guide from the VMWare site in order to make sense of what was submitted to them.

This kind of rush is not rare in the history of specification submission, even those that have been in the work for a long time . For example, the initial CBE submission by IBM had “IBM Confidential” all over the specification and a mention that one should retrieve the most up to date version from the “Autonomic Computing Problem Determination Offering Team Notes Database” (presumably non-IBMers were supposed to break into the server).

If lack of time is the main reason why all these APIs do not factor out the protocol aspects then I have no problem, there is plenty of time to address it. But I suspect that there may be other reasons, that some may see it as a feature rather than a bug. For example:

2) Anything but WS-*

SOAP-based interfaces (WS-* or WS-DeathStar) have a bad rap and doing anything in the opposite way is a crowd pleaser (well, in the blogosphere at least). Modularity and composition of specifications is a major driving force behind the WS-* work, therefore it is bad and we should make all specifications of the new REST order stand-alone.

3) Keep it simple

A more benevolent way to put it is the concern to keep things simple. If you factor specifications out you put on the developer the burden of assembling the complete documentation, plus you introduce versioning issues between the parts. One API document that fully describes the contract is simpler.

4) We don’t need no stinking’ protocol, we have HTTP

Isn’t this the protocol? Through the magic of REST, all that’s needed is a resource model, right? But if you look in the specifications you see sections about authentication, fault handling, long-lived operations, enumeration of long result sets, etc… Things that have nothing to do with the resource model.

So what?

Why is this confluence of model and protocol in one specification bad? If nothing else, the “keep it simple” argument (#3) above has plenty of merits, doesn’t it? Aren’t WSDM and WS-Management just over-engineered?

They may be, but not because they offer this separation. Consider the following practical benefits of separating the protocol from the model:

1) We can at least agree on one part

Thanks to the “REST is the new black” attitude in Cloud circles, there are lots of commonalities between these various Cloud APIs. Especially the more recent ones, those that I think of as “second generation” APIs: vCloud, Sun API, GoGrid and OCCI (Amazon EC2 is the main “1st generation” Cloud API, back when people weren’t too self-conscious about not just using HTTP but really “doing REST”). As an example of convergence between second generation specifications, see for example, how vCloud and the Sun API both use “202 Accepted” and a dedicated “status” resource to handle long-lived operations. More comparisons here.

Where they differ on such protocol matters, it wouldn’t be hard to modify one’s implementation to use an alternative approach. Things become a lot more sensitive when you touch the resource model, which reflects the actual capabilities of the Cloud management infrastructure. How much flexibility in the network setup? What kind of application provisioning? What affinity/anti-affinity control level? Can I get block-level storage? Etc. Having to implement the other guy’s interface in these matters is not just a matter of glue code, it’s a major product feature. As a result, the resource model is a much more strategic control point than the protocol. Would you rather dictate the terms of a contract or the color of the ink in which it is printed?

That being the case, I suspect that there could be relatively quick and painless agreement on that first layer of the Cloud API: a set of protocol considerations, based on HTTP and REST, that provide a resource control framework with support for security, events, long-running operations, faults, many-as-one semantics, enumeration, etc. Or rather, that if there is to be a “quick and painless” agreement on anything related to Cloud computing standards it can only be on something that is limited to protocol concerns. It doesn’t have to be long and complex. It doesn’t have to be factored in 8 different specifications like WS-* did. It can be just one specification. Keep it simple, ignore all use cases that aren’t related to Cloud Computing. In the end, please call it MUR (Management Using REST)… ;-)

2) Many Clouds, one protocol to rule them all

Whichever Cloud taxonomy strikes your fancy (I am so disappointed that SADIST-PIMP hasn’t caught on), it’s pretty clear that there will not be one kind of Cloud. There will be at least some IaaS, some PaaS and plenty of SaaS. There will not be one API that provides control of them all, but they can share a base protocol that will make life a lot easier for developers. These Clouds won’t be isolated, developers will use them as a continuum.

3) Not just one access model

As much as it makes sense to start from simple and mostly synchronous operations, there will be many different interaction models for Cloud Computing. In addition to the base operations, we may get more of a desired-state/blueprint interaction pattern, based on the same resource model. Or, somewhere in-between, some kind of stored execution flow where modules are passed around rather than individual operations. Also, as the level of automation increases you may want a base framework that is more event-friendly for rapid close-loop management. And there are other considerations involved (like resource monitoring, policies…) not currently covered by these specifications but that can surely reuse the protocol aspects. By factoring out the resource model, you make it possible for these other interaction patterns to emerge in a compatible way.

The current Cloud APIs are not far away from this clean factoring. It would be an easy task to extract protocol considerations as a separate document, in large part due to the fact that REST prevents you from burying the resource model inside convoluted operation semantics. To some extent it’s just a partitioning issue, but the same can be said of many intractable and bloody armed conflicts around the world… Good fences make good neighbors in the world of IT specs too.

[UPDATE: Soon after this entry went to "press" (meaning soon after I pressed the publish button), I noticed this report of a "REST-*" proposal by Mark Little of RedHat/JBOSS. I will reserve judgment until Mark has blogged about it or I have seen some other authoritative description. We may be talking about the same thing here. Or maybe not. The REST-* name surprises me a bit as I would expect opponents of such a proposal to name it just this way. We'll see.]

[UPDATE 2009/9/6: Apparently I am something like the 26th person to think of the "one protocol/API to rule them all" sentence. We geeks have such a shallow set of shared cultural references it's scary at times.]

[UPDATED 2009/11/12: Lori MacVittie has a very nice follow-up on this, with examples and interesting analogies. Check it out.]