HPC

Researchers boot one million Linux kernels as virtual machines

In a feat of Linux strength, computer scientists at Sandia National Laboratories in Livermore, Calif., announced that they had run more than a million Linux kernels as virtual machines. Previously, researchers had only been able to run up to 20,000 kernels concurrently. The scientists used virtual machine (VM) technology and its Thunderbird supercomputing cluster for the demonstration.

The aim of the project is to model malicious botnets, which are often difficult to analyze because they are geographically spread all over the world, explains Sandia’s Ron Minnich. The more kernels that can be run at once, said Minnich, the more effective cyber security professionals can be in combating the global botnet problem. “Eventually, we would like to be able to emulate the computer network of a small nation, or even one as large as the United States, in order to virtualize and monitor a cyber attack,” he said.

Running a high volume of VMs on one supercomputer — at a similar scale as a botnet — would allow researchers to see how botnets work and explore ways to stop them in their tracks. “We can get control at a level we never had before,” said Minnich.
Continued »

Ganglia 3.1 enables custom cluster, grid monitoring

Ganglia, community partner to GroundWork Open Source, releases cluster monitoring product Ganglia 3.1 at LinuxWorld Conference & Expo this week in San Francisco. Ganglia is a distributed monitoring system for high-performance computing systems such as clusters and grids. The central feature of Gangila 3.1 is that it allows administrators to create customized “metric modules.” Admins can create a module from metrics for CPU, network, disk and memory that they select à la carte, allowing for a tailored monitoring environment.

“I would hope that [Ganglia 3.1] changes business practices for the better, making clusters easier to use and more expandable,” said Ganglia developer Brad Nicholes. “We want to make sure that whoever needs to monitor data has the resources they need to do so.”

Previously, an administrator could create a metric module but could not integrate it into the Ganglia interface.Ganglia 3.1 allows an administrator to expand a cluster by adding custom metric modules on an as-needed basis. Ganglia 3.1 uses the round-robin scheduling algorithm, which enables admins to tailor the collected data to company’s needs.

Nicholes noted that it is important to upgrade all gmon agents, tools which allow a GUI to “talk” to the various components of a cluster, at the same time.

If you would like to use Ganglia with GroundWork Open Source’s GroundWork Monitor, GroundWork offers a Ganglia Integration Module that allows Monitor to provide multiple role status views, dashboards, reports, notifications and configuration tools.

Red Hat/Groundwork faciliate university’s high-performance computing

At the University of Utah, scientists who require access to computational resources rely on the Center for High Performance Computing (CHPC), a facility that maintains 872 nodes across three clusters. To run those clusters, the CHPC relies on open source software, specifically Red Hat Enterprise Linux as an operating system and GroundWork Monitor for IT and network monitoring.

Tom Ammon, a network engineer at the CHPC said that his organization has used GroundWork for nearly two years. The monitoring tool replaced Nagios, an open source network monitoring tool. “Nagios is really powerful and flexible, but it has an arcane syntax and it’s a hassle to configure,” Ammon said. The university’s switch to GroundWork was made with the hopes of reducing the time spent administering the monitoring system.

While GroundWork has been successful in terms of streamlining administration tasks, Ammon said that the tool has also brought greater functionality than he anticipated. “GroundWork is open enough so that we’ve been able to integrate it with a lot of open source packages,” he said. For example, the CHPC has integrated GroundWork with Cacti, an open source networking graphing tool. According to Ammon, combining GroundWork’s network status information with Cacti’s graphing capability enables CHPC to look at trending data. “We use trending data in the data center to monitor things like temperature and humidity,” Ammon explained. “It’s not that helpful to know what the temperature is right now if you didn’t know what it’s been like for 24 hours,” he said.

With Red Hat Linux running on clusters and open source tools helping to monitor and manage the operational work involved in providing high-performance computing resources, open source, said Ammon, “has really revolutionized the way we are accomplishing our mission.”

Linux still dominates the HPC arena

For some time now, Linux has been the dominant operating system in high performance computing. For everything from IBM, with its rockstar status supercomputer Blue Gene, to NEC or U.S. HPC players SGI and HP–the bulk of the leading HPC clusters today are Linux-based.

Four of the top five HPC systems in existence today are based on Linux, according to Top500 Supercomputing Sites, an independent web site that tracks the largest, fastest HPC deployments in the world. In 2005, when Top500.com started calculating which specific OS was dominating HPC, it found that Linux was used in nearly 80% of the world’s the fastest HPC systems.

The next TOP500 list will be released Nov. 13th (that’s tomorrow) during the Supercomputing Conference (SC07) in Reno, Nevada, but why wait until then for more Linux HPC goodness? UPDATE below: The list arrived early.

This morning our sister site SearchDataCenter.com broke the news that Sun Microsystems would be releasing two new systems designed to address the extreme computation, scale and storage requirements of today’s high-performance computing (HPC) customers. Called the Sun Constellation System the supercomputers are open computing environments that combine ultradense, high-performance compute, networking, storage, and software into an integrated “petascale” general-purpose system. Running Solaris, Linux and Microsoft Windows, the Sun Constellation System is designed to scale from departmental clusters to the largest supercomputer configurations, enabling customers to solve complex computational problems, the company said.

And that’s the first big Linux news of the day. Apparently, Constellation’s first user is the Texas Advanced Computing Center (TACC), which will put the finishing touches on a 504 trillion floating-point operations per second (teraflop) compute cluster in December. This CentOS Linux Cluster, named Ranger, will have 3,936 nodes, 123 terabytes of memory and 62,976 processor cores from AMD Opteron quad-core processors. All system components will be connected via a full-Clos InfiniBand interconnect. Eighty-two compute racks will house the infrastructure, which will sit in TACC’s J.J. Pickle Research Campus in Austin, Texas. IDC seems to think Constellation will make a pretty big impact in the HPC space. We’ll have to keep an eye out for it on future Top500 lists.

But wait, there’s more! According to 451 Group analyst and CAOS Theory blogger Jay Lyman, last week supercomputer superpower Cray rolled out its super scalable XT5 supercomputer and wouldn’t you know it, the thing runs Linux. It’s apparently the company’s biggest use of Linux to date, with Cray using AMD quad-core processors in a configuration of more than 1,000 CPUs to top 40 teraflops of performance. “As far as I know this is the highest density of Opterons you can buy in a system,” said Jan Silverman, senior vice president of corporate strategy, in an interview with EFTimes.com.

But let’s not forget about Microsoft’s HPC endeavors in the midst of all this Linux HPC love. Microsoft’s Windows Computer Cluster Server (CCS) is designed for the lower end of the HPC market, and industry watchers said the technology has been well received, particularly for small computing clusters. Vendors too have jumped onto the Windows CCS bandwagon, including HP, which extended a multimillion-dollar investment agreement with Microsoft to drive HPC into the mass market. It is also selling CCS 2003 as part of its HP Unified Cluster Portfolio, and Fujitsu Computer Products of America Inc., which recently published a best-practices paper for HPC cluster deployment, is using Microsoft Windows instead of Linux. But I’m still willing to bet that Linux has a lock on the list tomorrow. Call it an educated hunch.

Like I said, the list hits tomorrow at SC07. Let the chips fall where they may (and that’s a lot of chips — lolz!)

UPDATE: Or we could just have the list today. It’s out at 2:45 p.m. EST. Blue Gene again takes the top spot, but apparently there were plenty of surprises. More below… Continued »