Slashdot Mirror
Linux Kernel 2.6.11 Released
Xpilot writes "Linus Torvalds has just announced the availability of the newest Linux kernel release, 2.6.11. The newest addition to Linux that's stirring up some excitement is the inclusion of Infiniband support. You can get it from the usual mirrors at http://kernel.org/mirrors."
...before the site is slashdotted...
The Linux world is bracing for the final release of the new Linux 2.6.11 kernel, which will include a long list of driver updates and patches, with InfiniBand support perhaps being one of most interesting new additions.
Late last night, Linux creator Linus Torvalds issued the fifth release candidate for the 2.6.11 kernel. The first 2.6.11 RC was issued on Jan. 12; the second on Jan 21; the third on Feb. 2; and the fourth on Feb. 12.
In the RC5 posting, Torvalds indicated that it was likely the last RC before the final release.
"Hey, I hoped -- rc4 was the last one, but we had some laptop resource conflicts, various ppc TLB flush issues, some possible stack overflows in networking and a number of other details warranting a quick -- rc5 before the final 2.6.11," Torvalds wrote.
"This time it's really supposed to be a quickie, so people who can, please check it out, and we'll make the real 2.6.11 asap."
The long list of updates in the 2.6.11 kernel includes architecture updates for x86-64, ia64, ppc, arm and mips, as well as updates to ACPI (define), DRI (Direct Rendering Infrastructure, which permits direct access to graphics hardware for X Window System users), ALSA (Advanced Linux Sound Architecture, which provides MIDI and audio functionality to the Linux), SCSI (define) and the XFS high-performance journaling filesystem.
The 2.6.11 kernel will also be significant in that it includes driver support for the InfiniBand (define) interconnect architecture. InfiniBand, which is derived from its underlying concept of "infinite bandwidth," is a switched fabric interconnect technology for high-performance network devices that is common in a number of supercomputer clusters.
The upcoming inclusion of InfiniBand support in the Linux kernel is a major step according to the InfiniBand Trade Association.
"The inclusion of InfiniBand drivers in the upstream Linux kernel is a significant milestone," Ross Schibler, CTO of InfiniBand vendor Topspin Communications, told internetnews.com.
InfiniBand support was available previously in various Linux distributions, but it wasn't part of the mainstream kernel.org Linux.
"This now means that anyone that downloads a kernel will have automatic access to the software," explained Schibler. "It also means that any upcoming distributions (Red Hat, SUSE, etc.) will have the software included on their CDs. Previously SUSE had it on a distribution, but only in the 'unsupported' directory."
Schibler sees the inclusion of InfiniBand as a testament to the maturation of the technology.
"Now that the technology has matured to such a point that Linus has accepted it into the kernel, the way is paved for greater distribution of the code and accelerated deployment of the technology," Schibler said.
The previous Linux kernel.org release, version 2.6.10 was issued on Dec. 24 after two release candidates. Linux distribution began including the 2.6.10 thereafter with Red Hat's Fedora Project being one of the first.
Fedora Core 3 initially shipped with the 2.6.9 kernel and then upgraded to the 2.6.10 kernel on Jan 13. Mandrakelinux's 10.2 Beta 3 also includes the 2.6.10 release. SUSE Linux 9.2 currently includes the 2.6.8 kernel.
Including the most recent kernel into a distribution is not a particularly easy task. The upcoming Debian, code-named Sarge, will only ship with the 2.6.8 kernel. In a release update e-mail, Debian Sarge release manager Andreas Barth related that a meeting was recently held to review the status of which kernel they would include.
"The team leads involved eventually decided to stay with kernel 2.6.8 and 2.4.27, rather than bumping the 2.6 kernel to 2.6.10," Barth wrote. "This decision was made upon review of the known bugs in each of the 2.6 kernel versions; despite some significant bugs in the Debian 2.6.8 kernel tree, these bugs were weighed against the additional delays that a kernel version bump would introduce in t
Nigel: "You're up to 11 on your server and up to 11 on your network and you need that extra little push to set it over the top and you got nowhere else to go. So this amp goes to infinity."
Interviewer: "I see. But why not just make 11 louder?"
Nigel: "But it goes to infinity."
Interviewer: "Yeah, but what if you just made 11 as loud as infinity?"
Nigel: "But...it goes to infinity."
Google knows all.
Intell talks about Infiniband Architecture
Initially InfiniBand Technology will be used to connect servers with remote storage and networking devices, and other servers. It will also be used inside servers for inter-processor communication (IPC) in parallel clusters. Customers requiring dense server deployments, such as ISPs, will also benefit from the small form factors being proposed. Other benefits include greater performance, lower latency, easier and faster sharing of data, built in security and quality of service, improved usability (the new form factor will be far easier to add/remove/upgrade than today's shared-bus I/O cards).
Additionally, InfiniBand Architecture reduces total cost of ownership by focusing on data center reliability and scalability. The technology addresses reliability by creating multiple redundant paths between nodes (reducing hardware that needs to be purchased). It also moves from the load-and-store-based communications methods used by shared local bus I/O to a more reliable message passing approach.
Scalability needs are addressed in two ways. First, the I/O fabric itself is designed to scale without encountering the latencies that some shared bus I/O architectures experience as workload increases. Second, the physical modularity of InfiniBand Technology will avoid the need for customers to buy excess capacity up-front in anticipation of future growth. Instead, they will be able to buy what they need at the outset and 'pay as they grow' to add capacity without impacting operations or installed systems.
zanders at nvForums has posted patches to improve performance with 2.6.x kernels. Here's the thread:4 6676
t achmentid=10558
http://www.nvnews.net/vbulletin/showthread.php?t=
This is the cumulative patch:
http://www.nvnews.net/vbulletin/attachment.php?at
>InfiniBand, which is derived from its underlying concept of "infinite bandwidth," is a switched fabric interconnect technology for high-performance network devices that is common in a number of supercomputer clusters.
So that works only for supercomputer clusters ?.Interestingly, the ChangeLog has some very small number of entries. The one I found most fun was:-
I mean, other wise they would end up as "-1" or "0" (when you assume in code that "0" or "1" for 1 -bit fields). How did a sign-extension in the IDE (must be heavily used) be missed till version 2.6 ??. Typically, this looks like the average release - some bug fixes and a couple of big features which nobody (well almost nobody) would use on their boxes.Quidquid latine dictum sit, altum videtur
Yeah, all those people waiting to boot their supercomputers using Knoppix to take advantage of Infiniband will just have to keep waiting.
It's a rift in the time-space continuum.
Virginia Tech used Infiniband to wire up their G5 cluster. It's basically very fast I/O with some good logic built-in - "The technology addresses reliability by creating multiple redundant paths between nodes (reducing hardware that needs to be purchased)." is basically the same as the change from linear, Token Ring-ish networks to big Ethernet meshes like the Internet. I don't claim to know much at all about this, but seemingly it's the superior alternative today, and it sounds like it should be as well.
Linux 2.6.11 SCO 0 Better luck next time
For those of you that are unaware, since the poster doesn't explain at all what Infiniband is, I will explain it for you.
Infiniband is a high-speed, low-latency interconnect used heavily with beowulf clusters (currently). Infiniband, like Myrinet, addressed many of the problems that are inherent with using interconnects like ethernet.
The biggest problem with any TCP/IP based transport, in the world of supercomputing, is latency. The amount of error checking that is involved creates latencies that bring fine-grained (lots of memory reads/writes/swaps) calculations to their knees. As many clusters use MPI (Message Passing Interface) for sharing memory between nodes, a low-latency interconnect was needed to replace ethernet and TCP/IP. People have worked on reducing latencies over ethernet by designing raw transport stacks, relying on the switch and the quality/brevity of the ethernet connections (using short, shielded cables proved useful), to ensure accurate data transport, but none of these methods have proven viable.
Infiniband has also been used as an interconnect for network storage devices as there are obvious advantages to this; eliminating much of that latency makes reads and writes to a device much simpler thus reducing overhead and improving overall throughput.
More information on Infiniband can be found here at the Infiniband sourceforge page. This should give a sufficiently technical overview of what it does without any of the marketing talk.