Entries filed under “System Management”

As members of the HPC community, we often find ourselves needing to communicate the value of high performance computing to management teams who don’t quite “get it.” How we can do that more effectively is the subject of an upcoming insideHPC webinar sponsored by X-ISS, a leader in HPC implementation services.

Webinar Title: How to Talk to your CFO about HPC: Measuring Profitability and ROI
Date: January 22, 2013
Time: 8:30am PST

Panelists:

• Merle Giles, NCSA
• Sharan Kilwani, Industry Expert
• Ramesh Krishnan, ATK Aerospace Systems

Register Now

To set the stage for this topic, I caught up with Deepak Khosla, founder and CEO of X-ISS.

insideHPC: It seems like the HPC community doesn’t always speak the same language as enterprise executives, who seem to be focused on cost and ROI.
With all the broad cloud adoption going on, why do you think that cloud is gaining traction in the business world while HPC seems to remain flat?

Deepak Khosla: What we see in HPC is that everything is about Performance. And while we are seeing growth in both the HPC market and the cloud space, in many cases the cloud is a better fit for the Enterprise, which is where we are seeing more growth. There are more issues limiting HPC growth in the cloud, such as:

• Security –this is a common issue with both HPC & Enterprise, and the perception about data security uncertainty limits growth in the cloud.
• Big Data Sets – We see in HPC there often are very large data sets, and due to the bandwidth and latency issues in the cloud, access to that data with acceptable performance becomes an issue.
• Proximity – With most cloud implementations, there isn’t a guarantee of how ‘close’ the machines can be to each other. Low latency, high performance networks are highly utilized for in-house HPC environments today – but this isn’t always easily available with cloud environments. Even with 10 Gig technology, you have better bandwidth, but latency is still a concern. This leads to lower performance in clouds.
• Virtualization – whether it’s based on legitimate performance issues, or just tradition, many HPC sites apps tend to run better on hardware vs. virtual servers. Enterprise applications in general have been running on virtual environments for quite a while.

In the end, quite a bit is about performance – and ROI. If users can achieve the performance they need and address the above concerns, the cloud will become more widely adopted in HPC. But there is a break-even point in which having your own cluster system makes more sense too.

Finally, because X-ISS works with a diverse set of customers, we also agree with the model that this industry can be broken into two categories:

1. High Performance Technical Computing, or a more traditional HPC user; and
2. High Performance Business Computing.

We’re seeing growth in both categories, but High Performance Business Computing seems to be growing faster and since it is closer to the Enterprise ‘production’ business, it will be interesting to see the variance in cloud adoption rates between the two.

insideHPC: On your site, you talk about how there is a major shortage of skills and software available to provide management, analytics and ROI on HPC systems. What does X-ISS do to help mitigate this problem?

Deepak Khosla: These are two different problems that the HPC industry is facing. When you look at the issue of shortage of qualified people with HPC management skills – this has been an ongoing problem for years. Every organization is focused on keeping their HPC system tuned, running, and highly available to the users and this issue impacts their ability to produce timely results. X-ISS offers a turnkey outsource system management service called ManagedHPC that takes care of that challenge, by providing experienced people, and the right monitoring and reporting systems. This is a cost-effective solution especially for those sites where skills are not available or where running efficiently with best practices is important.

Regarding the shortage of analytics software, we see today that most commercial options are just providing technical analytics, and these solutions are also tied to a single vendor stack. Not only is this insufficient, it also sets up the undesirable vendor-lock. The reality is most HPC systems are very heterogeneous, made up of a variety of hardware and software stacks. Through our DecisionHPC software, we provide both business & system analytics on diverse systems, even geographically dispersed environments. The software was built to help improve performance, identify the overall cost, and to help organizations create strategies to track and allocate costs, thus understanding all of the performance and ROI implications.

insideHPC: Your DecisionHPC product is all about providing visibility into what is really going on inside the datacenter. What are your clients able to do with that information in terms of business advantage?

Deepak Khosla: At the high-level, accurate information is powerful. The ability to make changes and quickly react to improve execution, be more competitive, and more timely, is invaluable today.

For example, some organizations are in the business of charging for HPC cycles. Clearly, if this is your business model, then you need to know how past jobs were done, so you know how to price effectively. With this information you can react in a timely manner and improve your ability to perform successfully.

Or if you want to be able to charge back for projects to internal or external customers, you should be able to easily do that in a heterogeneous environment, with the right tools.

If high performance computing is an important aspect to your business success, it’s vital to have the information on both system costs and performance to maximize your HPC investment. It’s necessary to have the insight into how your systems and applications are performing and producing results in a production environment. These are business metrics like any other Key Performance Indicators – and better tools are needed to help businesses track, measure, and make better decisions today.

insideHPC: Large-scale clusters have existed for years. What is changing and why is it more important than ever to look at system management in a new light?

Deepak Khosla: One important change that we discussed earlier is that HPC is moving more and more into the business/commercial space. Here cost management, improving efficiencies, and productivity have a direct impact on business. Therefore it becomes vitally important to have the technical and business historical information; some of this has not been as important in the research space in the past. However, even in traditional HPC environments, there is more pressure to measure, track, report on, and make better decisions, based on the analytics created.

insideHPC: How does X-ISS turn instrumentation into business insight?

Deepak Khosla: Firstly we are now able to collect data from disparate sources – not just hardware, but also schedulers and other sources — and then we can associate those to business reasons, which allow us to formulate real business cost or performance. In other words, X-ISS can help customers identify and analyze how projects or business endeavors are performing, and the impact to the business.

Secondly, we also allow for the creation of business metrics, based on unique business needs. Because every business is different, the ability to track, store, and visualize these custom metrics lead to better insights, allowing for better decision-making.

insideHPC: Why sponsor a panel like this that focuses on how to talk about HPC to Execs?

Deepak Khosla: HPC is now being leveraged for commercial uses – more than ever before – this trend will grow. The purpose of this panel discussion that addresses the ‘Business of HPC’, is to help educate technical professionals, such as computer scientists and engineers, on how to have a dialog that relates to the CxO world, and how to identify solutions that can facilitate this dialog. An important part of this dialog are relevant metrics – how do we have a conversation on both technical and business metrics critical to success? HPC in the past, has relied on metrics more related to pure system performance and hardware and software purchase costs and not necessarily business performance. While many of the traditional tools used in research environments are not adequate for the next generation HPC use, there are solutions to this changing need. These new analytics tools are available today to help them manage HPC within metrics and KPI’s more associated with commercial businesses. The reason for this panel is our industry is ready to have this conversation.

We see from our perspective there is a growing need to treat all HPC systems as a business – and the language of business is numbers. This panel will open the discussion up, as to how a technical research and development manager, or research computing professional, can speak successfully to a CFO, or a committee evaluating ROI, and performance of any given HPC program.

Whether you’re fighting for grants, or fighting for internal funding, or fighting for customers, it’s time our community understands what metrics are necessary to be successful.

Register Now for the Webinar to be eligible to win a $100 AMEX gift card.