Virtualization Technology News and Information
Q&A: ScaleMP's CEO Shai Fultheim Talks Virtualization for Aggregation with VMblog

I recently had the opportunity to speak with Shai Fultheim, founder and CEO of ScaleMP, to find out more about what the company is offering and to find out more about the idea behind virtualization for aggregation.  And in this interview, we did just that.

1. VMblog:  Is server virtualization for aggregation just for HPC?

[Shai Fultheim] Aggregation provides a virtual machine with the combined power of a collection of servers.  For example, aggregating 10 systems will result in a virtual machine with an order of magnitude of more memory and CPU power, compared to physical servers.  

Any application requiring such compute capability can benefit from virtualization for aggregation, including technical computing, finance, analytics and data-warehousing.

An additional benefit of aggregating multiple systems: virtual machine operation simplicity.  The aggregated virtual machine requires only one OS, one storage pool and one jobs queue to manage.  This capability removes quite a bit of the complexity in running multiple physical servers.

The value of single point of management, however, is not related to specific application segment and applicable for any IT organization.

2. VMblog:  What are the types of applications currently needing SMPs and how do you see that changing?

[Fultheim] There are 3 application domains that require a large VM:

  • Applications requiring large memory
  • Applications requiring lots of CPUs in a shared-memory environment (multi-threaded applications)
  • I/O-intensive application

The explosion of cores per socket changes the way programmers leverage computers, and provides great simplicity, since many CPUs now have shared memory.  But programmers want to program for shared memory with no limits - beyond what's available within the servers.  Virtualization for aggregation fills that need -- with no limits - and we will see that growing in the years to come.

I/O-intensive applications such as data-warehousing and analytics can use large memory as a bypass for I/O.  We see this segment growing today, and I expect it will continue.

3. VMblog:  What are some of the issues that arise with traditional dedicated SMPs and how does server virtualization for aggregation work in comparison?

[Fultheim] Primarily, the issue with traditional systems is that one has to plan for peak load for the period (about 3 years) the system will be used.  Thus, these expensive systems remain underutilized for a significant portion of the system's lifetime.  

With aggregation that doesn't happen: you pay as you grow,  starting small and adding systems as needed. In addition, traditional systems, even the most recent ones, such as SGI's Altix UV system, run on slow processors.  With virtualization for aggregation, customers aren't locked in; they can choose the fastest servers and gain over 20% faster processors -- eliminating the compute bottleneck of traditional systems.
4. VMblog:  There's been a lot of talk of infrastructure as a service, how does server virtualization for aggregation enable this?

[Fultheim] IAAS allow customers to shape server resources to their application, and not be locked into their hardware's limitations.  Quite a bit of IAAS is provided using virtual machines, which segment a physical server into multiple, stand-alone entities - providing great ROI for service providers and customers.

With virtualization for aggregation, customers have more choice!  They can ask for a virtual machine with practically unlimited memory size or processors. Server aggregation without boundaries is the next wave of virtualization - and we start seeing it addressed by service providers.
5. VMblog:  What's new with vSMP Foundation?

[Fultheim] With vSMP Foundation 3.5 we provide enhancements in 3 key areas:

  • For data-center customers looking for on-demand virtual-machines, we now offer a greater choice of provisioning systems, with the integration of Bright Cluster Manager.
  • For customers seeking large-scale SMP, we offer up to 4x better performance on 4-socket systems, compared to the previous release.  In addition, we now support 4-socket systems from Dell, HP and IBM - providing more choice for customers.
  • For SMB customers looking to simplify cluster environment by aggregating cluster nodes into a single system, we offer a seamless MPI offload engine, delivering greater performance for distributed applications.

6. VMblog:  What are the drawbacks of cluster configurations and how does server virtualization for aggregation solve that?

[Fultheim] Clusters do not fit all customers.  There are quite a few customers who do not have the IT skill sets to optimize cluster deployments to run at optimum performance.  Many of these customers lack expertise in running distributed storage, and have no desire to spend lots of time on InfiniBand configuration.  Some IT organizations we work with have the skill sets to do all of the above, but are continually looking for ways to reduce datacenter operation costs.

That's where virtualization for aggregation comes into play.  Whether the end-user lacks skills to run clusters efficiently or whether the IT organization wants to reduce costs - aggregating all the cluster nodes to a single system makes the problem go away. With such a single system, end-users can focus on productivity and the IT organization need only run one server, instead of multiple servers.

Once again, I'd like to thank Shai Fultheim, founder and CEO of ScaleMP for speaking with me and answering a few questions.

Published Tuesday, November 09, 2010 5:34 AM by David Marshall
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
<November 2010>