RealTime IT News

Red Hat Expands Real Time Linux Cloud

There's fast and then there's real-time fast.

Linux vendor Red Hat (NYSE:RHT) is out today with its newest Real Time Linux platform, MRG 1.1 boasting new performance, messaging and grid computing (cloud) capabilities.

The new MRG 1.1 platform marks the debut of Red Hat's commercially-supported grid technology, which helps users create their own enterprise clouds as well as leverage the power of Amazon's EC2 service.

Red Hat is also claiming significant performance gains with a new Real Time Linux kernel and improved messaging speeds in a bid to appeal to unique sectors such as government, military and financial services.

Real Time is a technical ideal for providing deterministic performance. It enables actions within the same amount of time, every time -- a feature critical for a number of industries including telecommunications, the military, healthcare and financial services.

"MRG 1.1 is a pretty big update for Red Hat," Bryan Che, product manager for Red Hat MRG, told InternetNews.com. "A big theme for the release is formal support for grids. We previously had support for messaging and real time, but grid was in tech preview."

Red Hat announced the MRG 1.0 platform in December of 2007. With the 1.1 release, Che noted that Red Hat added a full Web-based console for grid management as well as the ability to offload to an Amazon EC2 instance as well.

A computing grid is a distributed utility computing environment where multiple node contribute their compute resources. It's a term that lately has become intermingled with the popular term Cloud computing, which also pulls on distributed resources for compute power. As for why Red Hat its Grid instead of Cloud? Target markets.

Che noted that traditional high performance computing environments, such as financial services and research are used to calling distributed computing a grid.

"They don't think of their own workloads as a cloud because they've been doing it for years," Che added. "We call it a grid but depending on who we talk to we emphasize the point its not so much it's a grid but it's the ability to execute any time of computation from sub-second calculations to persistent applications," Che said.

Another key improvement in the MRG 1.1 platform is improved low-latency messaging that can potentially impact performance with dramatic results. With MRG 1.1, Red Hat is including new RDMA (Remote Direct Memory Access) that can be used to take full of advantage of either Infiniband or 10 GbE (gigabit Ethernet) interfaces, Che said.

If you take 10GbE and just push it out over regular TCP/IP, latency will be improved over a 1 GbE connection though not as much as it could be. Che commented that hardware alone isn't enough to reduce latency; software needs to be optimized as well.

"So what we've done is written new drivers to take advantage of RDMA interfaces and this provides us with the ability to get orders of magnitude better latency than previously," Che said.

For example, Che explained that if you run messaging over regular TCP/IP you might get latency in the one millisecond range.

"On the other hand, once you use the RDMA drivers that we're introducing with MRG 1.1, we can drop messaging latency down to the 60 microsecond range," Che said.

The improvement in messaging latency capability could be a big benefit for Red Hat MRG customers like the US Navy as well as financial services clients who depend on messaging speed when executing trades.

Beyond the messaging latency improvements, Red Hat is also rolling in new Real Time Linux kernel improvements. Red Hat has been helping to lead the Linux community toward Real Time for the mainline Linux kernel since at least 2006. Che noted that work is still ongoing but the latest set of patches that will eventually end up in mainline Linux.org kernel releases are part of the MRG 1.1 release.

With MRG 1.1, tuning of the Real Time kernel itself gets a boost to further ensure that users maximize their infrastructure. A tuning tool call 'Tuna' gets a boost with new capabilities to create tuning scripts that can be replicated across multiple servers.

Real Time gives users deterministic behavior features, such that a given action will always occur within the same unit of time. With a non Real-Time system, behaviors do not have the same degree of determinism. Combining determinism with low latency messaging is a key value proposition that Red Hat is offering and it's one that that they claim has resonance in the market.

"The biggest interest come from customers that have very strong requirement with service level agreement or because they gain a competitive advantage by way of latency," Che explained.

"For example in financial services we see a lot of interest in the trading side where whoever has the lowest deterministic latency has competitive advantage. If you're submitting a bid and you incur latency, you're going to be beat up by the other traders that didn't have the latency so there is an arms race there."