More Thoughts on Cloud Identity

Unfortunately, I haven't had much time to blog lately, as November was essentially a wash. First, we were left without power for a couple of weeks by the epic Halloween storm that crippled much of New England early in the month. Then I took off on my annual pilgrimage to visit relatives in England. And then, of course, it was Thanksgiving. Next thing I know, the malls are filled with Christmas shoppers and the drone of those annoying seasonal tunes to which we are thankfully subjected for only a few weeks a year.

Anyway, I'm back now. While catching up on my favorite blogs, this comment on cloud-based identity by Sean O'Neill---who I consider to be one of the best minds in the business---caught my eye, mainly because it echoes my own sentiments on the topic.

Sean notes that from a technical perspective, there is nothing earth-shatteringly new about cloud computing itself, except that we now give it a fancy name. No argument from me there. But his more important point is that the increasing adoption of cloud services creates a whole new set of governance headaches for CIOs. To illustrate this point, he quotes the CIO of a major insurance company:

“One thing I have come to realize is that when I move my application to the cloud, all of the security of my networks and firewalls that I have invested in over the years disappears. The only defense I have left is identity and data security in the application”

In my experience, that sentiment is probably not unusual among CIOs and CISOs, particularly in highly regulated verticals such as financial services, pharmaceuticals and healthcare. Entrusting identity management to the cloud may seem like a good idea to analysts, vendors and techies, but they are not the ones who would be laying awake at night, worrying about the legal and regulatory implications of their cloud identity provider suffering a catastrophic breach.

As Sean explains:

Even if you can sue the pants off of your cloud provider, the basic problem is a breach would have occurred and your people are not involved at the security level.

In other words, if you are a CIO and sensitive personal information about your customers and employees is leaked due to a breach at a third-party identity provider, the victims aren't going to give you a pass because you entrusted security to a cloud service. If anything, they will hold you even more liable for gross negligence. Not to mention that "not my fault" will do nothing to mitigate damage to your company's brand reputation.

The increased adoption of cloud computing is inevitable, but it is both reckless and unrealistic to view IAM as just another one of those services, particularly for large organizations.

Currently, everybody seems to be focused on the idea of moving IAM tools to the cloud (which, by the way, does nothing to alleviate the process and governance issues that make IAM projects so notoriously complex---it simply moves these problems somewhere else). Instead of viewing IAM as just another commodity, we should be thinking about how to help organizations evolve robust governance strategies for managing cloud identities. I know this isn't as sexy as the idea of cloud-based IAM products, but it is far more relevant to the average CIO.

Is SPML Really Dead?

Over the past year or two, there has been an intense debate in the IAM community about the future viability of SPML. This began with a blog post by Burton Group's Mark Diodati in February 2010, in which he offered a bearish perspective on the poorly adopted provisioning standard.

I agree with many of the points Mark raised. Yes, SPML is too complex in its current incarnation. In trying to offer a broad range of functionality and account for every conceivable provisioning scenario, it often ends up accomplishing quite the opposite. Accordingly, SPML implementations tend to be extremely product specific, which defeats the entire purpose of a common standard in the first place. He is also correct that SPML implementations often cause performance issues due to the burden they place on connectors, although I would argue that this is less attributable to the standard itself than to how SPML has been implemented by provisioning vendors. And I completely agree with his assertion that enhancing the standard with a common inetorgperson-like user schema such as "SPMLPerson" would not only promote greater convergence with SAML, but would lead to increased adoption in the enterprise.

With that said, SPML works extremely well when implementations avoid connector-heavy and product-specific customizations (I'll come back to that shortly). In my opinion, the lack of SPML adoption isn't just because the standard itself is deficient, but because nobody has quite figured out the best way to implement SPML interfaces.

The most common use case I've encountered for SPML is when a customer wants to integrate their own proprietary UI with a commercial provisioning system. For instance, one of my customers has a sophisticated, AJAX-enabled access request portal built in .NET. It provides all the workflow functionality that one would expect to find in a commercial provisioning system, but was designed simply to manage requests and approvals rather than to automate provisioning. The customer had acquired Sun Identity Manager for provisioning automation, but didn't want to replace their elaborate, business friendly .NET interface with SIM's antiquated UI (and who can blame them?) In this particular instance, SPML was the obvious solution. The .NET application was modified to simply fire off SPML requests to SIM on the backend, thus automating provisioning actions while protecting end users from any visible impact.

In scenarios such as this, where a homegrown access request portal needs to integrate with a provisioning system, SPML works like a charm. But let's face it, this scenario doesn't require SPML specifically. In the use case I just described, Sun Identity Manager could have exposed any old SOAP or REST interface, and the integration would have been exactly the same, providing that the interface exposed the requisite methods.

The true value of a standard such as SPML is realized when providing interoperability with enterprise or cloud applications, and this is where SPML has unequivocally failed to deliver.

Which brings us back to the problem with SPML implementations. Because the standard is so broad and, unlike SAML, fails to define even a minimal schema for user profiles---much less offer a reference implementation---identity vendors have tended to implement variations of SPML that are product specific and are tightly coupled with the connector layer and underlying data stores. The result is that most SPML implementations are ugly and XML-heavy, which in turn can lead to performance issues when performing large numbers of provisioning transactions.

I have yet to see an SaaS application that supports inbound SPML messages in a way that could be called "standard". Likewise, I have yet to see a user provisioning solution that supports SPML in a non-proprietary fashion. As a result, provisioning solutions for cloud applications still depend heavily on individual connectors that all implement different product APIs, which is to say that SaaS applications are currently treated no differently from any other kind of application.

So the question remains, is SPML really dead? Well, in its current incarnation, it's hard to argue that it was ever really alive to begin with. SPML 2.0 is now five years old, and the OASIS Provisioning Services Technical Committee hasn't convened since 2008. But the critical need for interoperable provisioning is now greater than ever, particularly considering the explosive adoption of cloud applications in recent years. The fact that most identity management RFPs include a requirement for SPML confirms that standards-based integration remains a critical priority for organizations. Even as IAM maturation drives us away from traditional user-centric IAM towards more business-centric models, identity management projects still spend far too much time focusing on provisioning, and specifically on the nuances of proprietary connector architectures; this is another argument for a widely adopted provisioning standard that abstracts the complexity of underlying systems and data stores.

If SPML is to survive, then it will have to be reincarnated without any consideration to backwards compatibility. For that to happen, a major player in the identity space (I'm talking about the likes of Oracle, CA or IBM here) needs to step up, take leadership and create some momentum for a new standard.

In the meantime, there are several proposals out there to compensate for the lack of momentum surrounding SPML and the critical needs that it was intended to address. Jim McGovern, never a shrinking violet, has a wish list for what he would expect to see from SPML 3.0, if it ever happens. In 2009, the visionary Nishant Kaushik proposed using OAuth to support federated provisioning needs. The SimpleCloud (SCIM) initiative----which enjoys participation from various cloud providers such as Google, Salesforce as well as identity vendors such as UnboundID and SailPoint----appears to have embraced this concept and taken it to another level. For my money, SCIM is the most promising candidate for a new provisioning standard. Seemingly inspired by the LDAP standard, SCIM also leverages SAML, OAuth 2.0 bearer tokens, OpenID and JWT, while favoring a REST API over less efficient SOAP.

In summary, I tend to agree with Mark Diodati that SPML in its current form is obsolete, but the need for an interoperability standard is now greater than ever. Two lingering questions remain. Will SCIM be the standard we've been waiting for, and what is the value proposition that will compel vendors to embrace it?