This post continues the discussion in my “The Cloud Mystery Machine” post.
Contrary to the utopian cloud model many espouse, underlying infrastructure still matters, especially when you consider application performance and the ability to satisfy SLAs. Most folks are well aware of virtualization live migration incompatibilities between Intel and AMD platforms, but what about substantially reduced performance between Intel platforms? For example, consider an enterprise application that realizes a substantial performance benefit from Intel’s hardware-assisted memory virtualization – Extended Page Tables (EPT). XenApp, Exchange, and SQL are among the many applications that benefit from EPT. Moving the application to a cloud platform without EPT support could result in significantly degraded performance. If you have no information on a cloud service’s underlying hardware infrastructure, you may not know about the lack of EPT support until you start getting complaints about application performance. Then what do you do? Ideally, when a user or application requests infrastructure services from a cloud provider, there should be a mechanism for specifying low-level hardware requirements such as EPT.
When it comes to trusting critical applications to cloud service providers, not all organizations are keen on the idea of trusting an application on a white box server with memory acquired from the lowest bidder. Ditto for backend storage. This is one of the reasons why initiatives like VCE are important. A cloud provider offering services on known and trusted hardware is important to many decision makers. Sure you may pay a little more for the service, but for many IT decision makers, that’s OK. SLAs are only as good as a provider’s capability to honor them. Without knowledge of the provider’s infrastructure, you may be rolling the dice. Take the recent HostV failure as an example. While phrases such as “All you need to worry about is the SLA” sound good in theory, today they’re simply not practical. If cloud providers truly want to enable services such as application bursting to cloud, then they need to provide interfaces that accept very specific infrastructure requirements. Open Virtualization Format (OVF) is the best option we have for importing VMs to foreign infrastructures, and it’s extensibility allows room for custom metadata today. So there’s really no reason why service providers can’t offer such capabilities today. Also, there’s no reason why a service provider shouldn’t be transparent about their physical and virtual infrastructure. If a provider isn’t offering the detail you need to feel comfortable, then move on. There are plenty to choose from.