August 2, 2018 | By Benzi Galili
ScaleMP continues to expand the features and capabilities of its virtualizing and aggregation for scale-up x86 solutions, including aggregation of CPUs, memory and IO devices, support for accelerators and GPUs, memory expansion over storage-class memory, and certification for high-speed fabrics and high-end storage. We are helping our clients discover new efficiencies as well as innovative capabilities that just aren’t possible when constrained within a physical, hardware-only world.
Here are three great reasons why you should be implementing software-defined computing and server aggregation in your HPC infrastructure today:
1. Standardize your data center.
Stop buying specialized, proprietary and expensive platforms for your most demanding workloads. Instead, use ScaleMP’s vSMP ServerONE to create virtual scale-up servers that exactly match your application workload requirements. Use your budget wisely to buy the most cost-effective, commodity-off-the-shelf (COTS) servers, and use software to create a virtual server at the desired scale. Does your application need 4TB of memory? No problem! Is a user asking for a single server with 1,000 cores? Give it to them with the press of a button! This approach translates to a broader choice of IT vendors, more standardization in the data center, more computing power for lower budget and a smaller data center footprint — all of which translates to lower TCO and a more flexible and agile infrastructure.
2. Increase data center utilization.
With the ability to create server systems at virtually any scale from commodity hardware, using ServerONE software to scale up rather than specialized hardware ensures your data center utilization will increase. In the past, organizations purchased large, proprietary SMP machines, or “fat nodes”, for those jobs that had specific memory requirements or needed high core counts. But those days are long gone. It is a clear and proven law of economy: The more specialized the resources are for specific tasks, the lower the utilization. Indeed, in many cases, the SMP servers just sit there underutilized, or are used to run throughput or MPI jobs just to avoid the idle time. I’ve seen a customer use a two-year-old SMP —that was barely used 10 percent of the time since it was put into production — as a makeshift file server. With the ability to virtualize all the components of multiple servers and virtually reassemble them to match any given workload, large portions of your data center won’t sit idle; nor will high-end hardware be used to run low-end jobs.
3. Help move things to the cloud.
You didn’t expect to read an IT-related blog post without seeing the word “cloud” in there, did you? Think about it: You virtualize and software-define everything. You use containers and VMs to partition hardware when needed. You use software-defined storage and networking in order to better utilize and lower the TCO of your infrastructure. You strive to deliver systems that adapt to user workloads on demand. Moving away from dedicated hardware per application and virtualizing the infrastructure on which the applications run is the first step when preparing for the cloud. Be it a private or public cloud, you want your users to worry less about physical resources, and more about functionality and fit for application needs. By maintaining your legacy specialized SMP hardware, it will be a drag on your efforts to prepare for a future on cloud nine . ServerONE will allow your most demanding HPC workloads to reach for the cloud.
You are soon going to be engaged in a cloud HPC deployment, or you might already be involved in one. If your intention is to make your entire environment software-defined, don’t overlook your servers. By adopting ServerONE, you reap the benefits of standardization, increased utilization and on-demand provisioning, ultimately increasing your scalability and agility and reducing TCO.