Original URL: http://www.theregister.co.uk/2009/02/10/redhat_mrg_fedora11/

Red Hat updates real-time Linux

Fedora 11 goes alpha, too

By Timothy Prickett Morgan

Posted in Storage, 10th February 2009 01:31 GMT

Red Hat has announced that it has begun shipping the second rev and the first fully functional version of its Enterprise MRG real-time Linux. The Fedora Project, which is sponsored by Red Hat, has also put the alpha of its Fedora 11 development release in the field on time.

Enterprise MRG is a variant of the Red Hat Enterprise Linux stack that has been tuned to support messaging, real-time, and grid computing workloads rather than the standalone variant of Linux, which is tuned for file, database, and application serving. Enterprise MRG 1.1 was supposed to ship at the end of 2008, according to Red Hat's original plan, so it is a few months late.

Enterprise MRG replaces the generic Linux kernel with a real-time kernel based on the CONFIG_PREEMPT_RT patch set that was created by developers at Red Hat, Novell, IBM, Silicon Graphics, and a number of other vendors to make it more of a real-time operating system, one with a streamlined kernel offering the lowest possible latencies for applications. Novell's SUSE Linux Enterprise Real-Time, or SLERT, variant of SLES 10 is chasing the same market niche. The R in MRG is the part that financial services companies and defense contractors are interested in for their respective trading and weapons systems, where latency means money or death.

The M in MRG is for message queuing, and is designed for applications that pass lots of messages around between applications and servers, allowing relatively loose coupling of applications across servers. IBM's WebSphere MQ, Microsoft's Message Queuing Middleware, Tibco's Rendezvous financial transaction messaging, and the Java Messaging Service examples of the kind of messaging that Red Hat is talking about. But Red Hat's messaging within Enterprise MRG is based on the Advanced Message Queuing Protocol (AMQP), an open and standard implementation of messaging technology.

It is the G part of MRG that has been missing up until now, and that is because Red Hat's programmers have been integrating the Condor grid software developed by the University of Wisconsin into the Enterprise MRG product. When MRG went into beta in December 2007, Condor was not yet woven in and the release did not yet support the JBoss Enterprise SOA stack, either. Enterprise MRG 1.0 was launched in June 2008 as a commercial product, but the grid components were still only in technology preview.

With MRG 1.1, Red Hat has tweaked the real-time kernel so it works better on processor chips that have multiple cores, which is important as Intel and AMD move from two to four and soon to six or eight cores on processors aimed at servers.

MRG 1.1 also has native InfiniBand and Remote Direct Memory Access (RDMA) drivers for lower-latency clustering. The RDMA features, which works over InfiniBand or Ethernet networks (depending on the underlying hardware support for RDMA, of course), allow nodes on a cluster to move data between nodes directly into and out of memory in the nodes rather than having to go through the I/O stack. This really cuts down on latency.

The Condor grid software that Red Hat has been testing for the past eight months and is now in production allows users to create local or remove grids, using physical or virtual machines or a mix of both; it also allows HPC applications to steal cycles from desktops and laptops linked to the network (the original meaning of grid as it applies to IT and the one that most people remember). Red Hat said that the grid portions of Enterprise MRG can be used to dynamically provision cloud-style infrastructure, such as software stacks running on Amazon's EC2 grid. So customers could deploy Enterprise MRG on their own grids and, if need be, buy capacity on the public EC2 grid and provision the same applications they are running internally over on EC2.

Red Hat may be an open source software company, but it has been cagey about pricing on Enterprise MRG. The company said that it is aiming to keep the price of Enterprise MRG down to less than twice the cost of a standard RHEL license on a given box, and has said in the past that its goal was to price it to be competitive to the maintenance charges that customers pay for IBM's WebSphere MQ or Tibco's Rendezvous messaging software. Red Hat was planning to offer the real-time and messaging extensions to RHEL as standalone products, but the grid software requires the whole shebang.

While Red Hat Enterprise Linux is available on x86, x64, Power, Itanium, and mainframe servers, Enterprise MRG is only available on x86 and x64 servers, and considering that Intel and AMD don't ship 32-bit server chips any more and the code it is really tuned for x64 iron, the x86 support is kinda silly. But it doesn't cost anything to have it.

Red Hat and Novell have been duking it out to show whose real-time Linux has the lowest latency and most throughput. They keep leapfrogging each other in the benchmark tests that they are touting. Which is what competition is all about. It would be interesting to see how Solaris 10 stacks up to both Enterprise MRG and SLERT.

Fedora 11 goes alpha

Red Hat has also announced that the Fedora Project has put the alpha of the upcoming Fedora 11 release out, and on time. A month ago, the Fedorati had their user and developer conference, FUDCon in Boston and hammered out the new features they want to add.

According to Paul Frields, the Fedora project leader, the community is spending some time refining and enhancing the process whereby users can make "spins," or customized versions, of Fedora for their own particularly uses and to share with others. While Fedora 10 already has some tools for creating custom distros, the Fedora community wants to make it easier to make and share variants that might be tuned for specific languages or workloads.

Fedora programmers are also trying to get the system boot time down to under 20 seconds with Fedora 11 - Fedora 10 was 30 seconds. The KVM virtualization hypervisor, which is part of the mainstream Linux kernel and which Red Hat gain control over when it paid $107m for Qumranet last fall, is the default hypervisor on Fedora 11; the Xen hypervisor, which is still the default in Enterprise Linux 5, is an option on Fedora 11.

Fedora 11 is expected to go into beta on March 24, the preview release is expected to follow on April 28, and the final release is slated for May 26. ®