Does the cloud lend itself to a major all-in-one platform, or should enterprises pursue a piecemeal approach, at least for the moment?
It should be no surprise that the major platform vendors like Cisco and HP did what they do best for the burgeoning cloud movement -- build all-inclusive integrated infrastructures designed to literally remake the IT environment. And more than likely these behemoth systems will find their way into top-tier enterprises looking to hit the cloud running.
But there is also something to be said for leveraging existing infrastructure as long as possible, and many smaller companies are quick to point out that cloud services can be delivered, both internally and externally, with largely the same virtual infrastructure that is already in place.
One of these is ScaleMP, which this week unveiled the vSMP Foundation for the Cloud. It is essentially a server-aggregation tool that allows you to gather existing commodity servers, and the VMs they contain, into a symmetric multiprocessing (SMP) pool that can extend beyond the data center walls. What you get is the ability to provision and reprovision system resources according to data requirements. The system allows you to aggregate up to 16 x86 servers, which would provide 128 cores if you're using the newest Nehelems. It also provides seamless integration with most systems management and provisioning stacks and supports a variety of programming models, such as serial, throughput, multithreaded or large-memory.
It also represents ScaleMP's decision to head down-market from the HPC realm it has occupied for the past several years. As The Register points out, the system has the ability to convert four two-socket servers into an eight-socket cluster without an expensive InfiniBand switch to tie them all together. Instead, the system uses a proprietary interconnect called Direct Connect 2 (DC2) that uses InfiniBand adapters and drivers to create a virtual backplane. So not only do you save a few bucks off the IB switch, but you can generally pick up four two-socket machines for less than a single eight-socket.
Another company eager to get in on the cloud is Liquid Computing, which recently teamed up with Intel to support its LiquidIQ fabric-management software on the Server System SD1680MV rack server as well as NetApp storage units. The claim to fame here is that, unlike HP, Cisco and others, LiquidIQ is platform-independent, so it should fit in nicely with legacy heterogeneous infrastructure. The company promises a 10:1 reduction in provisioning time and cabling, and a 2:1 reduction in space requirements, provided you deploy the highest-density Xeons.
There are no doubt many other smaller companies angling to get a leg up on the cloud, and by nature they will be forced to adapt their technologies to existing infrastructure rather than the other way around.
Ultimately, though, as cloud deployment patterns start to take shape, speed of deployment and management simplicity will gain in importance, which is where the all-in-one platforms shine.
The question is: Once the inevitable platform wars are under way, with there be any room for both vendors and customers who want to go their own way?