2011 Prediction: ScaleMP
Gartner recently predicted – predictably – that cloud computing will be a top 10 strategic technology for 2011. Farther down their list they also included fabric-based infrastructure and computers, which they defined as “a modular form of computing where a system can be aggregated from separate building-block modules connected over a fabric or switched backplane.” Why will aggregating commodity systems be a top technology in 2011?
For decades, system manufacturers have been developing server systems in a myriad of form factors in order to best target different application workloads. Scale-up servers the size of a large cabinet for high end workloads, small blade servers for transactional or parallelized applications and deskside servers for smaller customers without dedicated datacenter environments. Due to the distinct characteristics of individual applications and workloads, each of these designs come with multiple variations and each system vendor supports and sells dozens of different systems, each optimized for a different use case. The biggest issue with having purpose-built systems is that there is a lack of flexibility. IT has to run a certain type of application on a certain type of system.
What we’ve seen recently is an increase in the number of cores that commodity servers have. For example, Intel has dual socket systems with 12 cores, and four socket systems with 32 cores. AMD systems packs the greatest number of cores, with 24 cores for dual sockets and 48 cores for four socket systems.This trend is great for a growing number of applications that require significant amounts of memory, I/O and processing power – applications in fields such as bioinformatics, finance, analytics and data warehousing. Organizations can use aggregation techniques to combine commodity servers for these computing intensive workloads.
One emerging technology, virtualization for aggregation, combines physical servers or even virtual machines (VMs) to create a large virtual symmetric multiprocessing (SMP) system. However, virtualization for aggregation use extends beyond the traditional high-performance computing needs. For example, one of the issues that will arise as servers increase their core count, is that when VMs are placed on such a server, the amount of memory per core decreases. One way to counter this issue is to aggregate a few VMs so that they can be managed as a single VM from a single management point.
VM on VM capability, as this technology is called, empowers end users to shape compute resources to fit their workloads on-the-fly. The first use case creates one large VM out of many small VMs – partitioning many servers into multiple virtual machines and then aggregating some of those VMs into a larger virtual SMP system, depending on need. This situation may occur when an IT manager has deployed multiple virtual servers taking up a certain amount of CPU power, but is also setting aside CPU space in case the need arises. By running a hypervisor for aggregation, an IT administrator can collect and aggregate all this available yet unused computing resource across the data center or private cloud to create a new larger or customized system that can be utilized for another workload.
The second use case places many small VMs on top of a large VM, which is essentially a large SMP created using a hypervisor to aggregate multiple nodes into a large virtual system and then splitting that system into multiple VMs using a hypervisor for partitioning. Provisioning smaller VMs from this larger pool provides a better platform for load balancing, dynamic system provisioning and hardware migration. Ultimately, IT can also dynamically add and remove resources to the pool.
