Entries filed under “Sponsored Post”

Univa Grid Engine is engineered and supported by the most celebrated gurus of the Grid Engine world–who are at the core of the Univa development and support team. After 18 months of development and three production releases later we asked our engineering team to talk about their experiences with Grid Engine and the new features they have been working so hard to complete at Univa. The result is a series of blog posts called “Engineers Talk” in which the team opens up and talks about their product. It is, after all, the people at the backend that make the difference.

To get a taste of what we mean, here is an excerpt from the blog posted by Ernst Bablick, a 13-year year veteran of developing Grid Engine. Ernst describes the painstaking process of developing the powerful new feature in Univa Grid Engine called “Job Classes.”

I am often astonished how difficult and long-lasting it can be to design and implement new features.” said Ernst Bablick, 13 year veteran Grid Engine developer now working at Univa. “Meanwhile I am working for more than 13 years in the team of engineers that is continuously improving and enhancing Grid Engine. About seven months ago I started with a functional specification document describing a new object type called job class. A month later I was able to start with the implementation. Now, 319 modified files, 21,737 insertions and 5,673 deletions later, I am in the middle of the test phase. Our team created new alpha packages a week ago and test results indicate that we can release job classes with Univa Grid Engine 8.1 soon. It’s the right time to explain job classes in more detail with this and the upcoming blog entries….”

The bloggers discuss the following new features in Univa Grid Engine 8.1:

• Job Classes Introduction
• NUMA Aware Scheduling
• Better Resource Management w/ RSMAP
• Simplified Debugging
• New Spooling Method
• Fair Resource Urgency
• Configured Parallel Environment Selection Order

To read the full set of posts, please visit our blog wheregridenginelives.com.

Hello,

Although you might be familiar with the solutions that ScaleMP offers to the HPC community, I’d like to take this opportunity to tell you more about our new product offerings and share with you my view about the datacenter of the future. With the growing demand is to handle larger jobs in less time, with tighter customer budgets, as well as financial difficulties and uncertainty affecting some traditional SMP vendors, now is a great time to investigate how you could increase the flexibility of your datacenter in order to provide for varying workload demands. To assist with all of the above, we are now offering our products for both Intel Sandy-Bridge as well as AMD Opteron 6200 (Interlagos) processors – delivering the highest performing shared-memory systems on the market. In addition, we recently announced a migration plan for SGI customers worried by SGIs recent financial announcement and I strongly urge you to look at our on-demand SMP offering.

Just as you would not consider buying a film-based camera anymore, we have reached to the age where virtualization software, much like digital pictures for photography, forms the future of computing. ASIC-based SMP systems are similar to film-based cameras. Kodak, the market-leader of the film-based cameras has filed for Chapter 11 bankruptcy. Similarly, systems vendors that develop ASIC-based SMP solutions are (again) experiencing financial challenges, requiring you to look for alternative solutions.

ScaleMP has been delivering software that creates virtual shared-memory systems (SMP or NUMA) for a number of years. We refer to our products as Versatile SMP™ (vSMP). Our software is flexible in that it allows you to create a larger, more powerful SMP systems from low cost components. By combining the resources of lower cost servers into a larger, easier to manage larger server, cost savings can be realized without sacrificing performance.

In addition to our traditional shared-memory offering, we provide the HPC community with:

• On-Demand SMPs – with vSMP Foundation for Cloud, SMPs can be created on a small portion of a large-scale cluster based on workloads. There is no need for dedicated, expensive SMP systems that cannot adjust to changing workloads. With vSMP Foundation for Cloud, as the workload changes, these systems can be returned to the cluster to be used as compute nodes. As user requirements dictate, large or small SMPs can be instanced, increasing the ROI.
• Cluster management simplification – what could be more effective to small cluster owners than having only one operating system installed and maintained for the entire cluster? Compare this with traditional clustering where multiple OS instances must be maintained, and fabric management and clustered file systems must be set and configured. With vSMP Foundation the entire cluster infrastructure is virtualized, so that system administrators do not have to become involved with its complexities – and can manage it just like a large workstation.

In summary, virtualization brings your HPC infrastructure performance, flexibility and cost savings. Whether you need a large-scale shared-memory system for your most demanding workloads in either a fixed or an on-demand-environment or whether you are looking to simplify your cluster infrastructure – virtualization has it all. Find out more about ScaleMP offerings by clicking here.

Shai Fultheim
Founder, President and CEO,
ScaleMP