This article is more than 1 year old

TurboLinux clustering and the code-forking scare

Or, the story that wasn't

TurboLinux airdropped some 'clustering enhancements' on to Linuxland this week, but it wound up mounting a search-and-rescue operation before anyone noticed. The kernel patch adds some load-balancing which TurboLinux says will make Linux more attractive as a clustering option. It's a patch that's optional, of course -- all modifications to the kernel have to be approved by Linus Torvalds himself, and Linus is a man who likes to say no -- having veered to the cautious side of most major decisions, not least SMP. The small patch itself is issued as GPL, but the TurboLinux business plan is to release the kernel bits as GPL code and the rest, running in user space, as proprietary binary-only extensions. Non-TurboLinux users will have to patch the kernel, write their own user space extensions, and hope for the best. Clearly, this is a strategy designed to gladden the hearts of their VCs. Now this unremarkable posting would have got little attention, but for the intervention of ComputerWorld and the usually reliable New York analyst outfit DH Brown Associates, who saw a chance to give us that old favourite -- 'when will Linux fork?' ComputerWorld, the puzzling IDG weekly which has jumped onto the Windows NT bandwagon just as the other trade weeklies are jumping off, [ADD YOUR COMPUTERWORLD HORROR STORY HERE] quoted DH Brown's Tony Iams as saying that Linus had better goddam include it or the kernel would fork. This might be a very good question -- if AT&T and Sun had wanted a unified Unix kernel badly enough fifteen years ago, and convinced DEC and IBM they needed one too, history would have been very different. For years, Microsoft has made much of NT being one code base (hmm... Embedded, Workstation, Data Centre, Server with Multiwin, Server without MultiWin, anyone?) And the point hasn't escaped the Microsoft spin-machine either, which has been briefing friendly journalists about the parallels for a while now. But it ignores a couple of facts, we reckon. One is that GPL forks are yet to produce a serious catastrophe: on plenty of occasions the code has merged back into the main tree -- as with the C libraries. The unyielding Torvalds has an uncanny knack for conflict resolution, and when it comes to the crunch the Linux herd likes to stomp as one. Or, else -- and this wasn't GPL, but the point holds true -- as in the case of the X11, the copyright holder simply gave up being bolshy and decided to go with the flow. And is Linus likely to hand running over to one distribution? TurboLinux seemed to realised this too last night. The distributor's source tree maintainer said that he'd never submitted the code to Linus, and didn't plan to. And that it was only ever a DiY patch, so if anyone wanted to get their hands dirty, they should feel free. Alan Cox, second to Torvalds in the kernel pecking order, tartly pointed out "Using Wensong Zhang's code [alternative load balancing] because it is rock solid and production hardened. It needs no proprietary tools. Several vendors already ship this code. I also know people building big web setups using it. [www.linuxvirtualserver.org]" So there. Now we're indebted to Mr Jesse Berst of ZDNet's Anchor Desk for pointing out that what gets Redmond really alarmed is the prospect of high availability Linux clusters that scale like hell. Despite trying to port 128-node VMS clustering to NT in the mid-90s in a vastly expensive programming gang-bang, using many of DEC's original engineers, NT clustering today supports, er…two nodes. Clearly there's something rotten. But the fab four of the self-styled Linux Cluster Cabal (Braam, McVoy, Scottish wunderkind Tweedie and the other one whose name we always forget) have been planning as much in not so secret fashion for a while now. We're expecting some kind of white smoke to emerge from the Cabal RSN -- so um, over to you, Mr Tweedie? ®

More about

TIP US OFF

Send us news


Other stories you might like