Amazon Web Services (AWS) this week announced that new C4 instances based on Intel Xeon E5-2666 v3 processors that Intel worked with AWS to optimize are now available. Because these are the fastest instances of AWS, interest in C4 will no doubt be high. But AWS is hardly the only cloud service provider building its own servers in the cloud. In fact, the prevalence of custom servers in the cloud is one of the primary reasons shipments of commercial servers now only grow in the range of single digits year over year.
Google is another example of a cloud service provider building custom servers. In addition to using x86 servers, Google is also starting to employ servers based on designs from the OpenPOWER Consortium, which IBM created to help fuel the development of Power processors as an alternative to x86 servers. Oracle, meanwhile, is making broad use of both Sparc and X86 processors across its cloud, and Verizon has deployed x86 servers based on processors from Advanced Micro Devices (AMD).
In addition, new classes of 32 and 64-bit ARM processors are starting to show up in the cloud. Even commercial server vendors such as Hewlett-Packard are recognizing this new reality. HP last year formed an alliance with Foxconn to jointly design and manufacture servers optimized for cloud service providers.
The significance of all this activity from an IT service provider perspective is that the notion that cloud compute has become a commodity is only half truth. While pricing for compute services in the cloud is comparatively low, the performance of different types of application workloads running on different classes of processors will vary widely. As such, IT service providers will need to be a lot savvier about where they actually decide to run application workloads in the cloud.
The rise of hybrid cloud computing
In fact, it’s that very issue that is driving new forms of hybrid cloud computing. While most of the hybrid could computing discussion today is focused on workloads that span on premise systems and servers in the cloud, it’s more than likely that bigger issue will be unifying the management of application workloads running on different cloud services. For example, a Big Data analytics application might be running on one cloud platform, while the ERP applications it needs to be integrated with are running on a cloud platform optimized to process transactions.
While a school of thought has emerged in recent years that says processors don’t much matter in the age the cloud, the fact is nothing could be further from the truth. Processor speed and the amount of memory they make available not only impact application performance, but the processors themselves are taking over management of the hypervisors that run on top of them.
It’s still too early to say exactly how all this will play out across enterprise IT. But the one thing that is certain is that going forward, cloud services will arguably be anything but a commodity.