Well, I knew the folks to code on the kernel are an elitist bunch, but I thought it was because they were god-like. Turns out that lanl Advanced Computing Lab folk have been finding a lot of outstanding bugs in various bits (bsd kernels too! but they often use linux b/c it offers the best SMP support while providing source to work with), and as they report to the kernel folk what is broken and suggest how it might be fixed, the kernel list is often quiet. The coders are seemingly ashamed of their code, and often try to explain that "it's not a bug"... This is ALL 2nd-hand to me, so chew yer salt. Anyway, I started by asking all about how linux is used and why they like it, assuming that "it's so great"... proceeding to get the pragmatist view of those who struggle to make broken bits work. They've got LOTS of great hacks going on, from shadow-processes of all cluster node processes on the head node, to patches to gcc for the G4's altivec. One thing that kinda bummed me out when visiting was that there's SO MUCH hardware inside the facility that's not being used. I kept asking "Why can't I just take the old outdated stuff? It'll make room for more taxdollars..." but they really have a problem with tax dollars leaving the lab : ) I just thought linux was somehow perfect for that [cluster] environment, but it is not SO different from any other OS. There is some Plan 9 momentum there, and the visualization cluster (I think the one recently slashdotted was not ACL's) runs windows... b/c linux never has the latest graphics hardware support. It made me sad to see the windows cluster; I was ROTFL at it for a bit, but it was a good lesson to see 'all the children playing nice together...' Just like in the Real World (no not mtv's) when systems are designed with Requirements, instead of just 'seeing if we can get it to run', as opposed to making an ad-hoc "cluster" out of virtual machines. It just blew my mind, everything that goes on the lab. The linuxbios folks are burning eeproms and breaking mobo's, and in the same room, inter-process communication and memory mgt is being worked out for some *massively* parallelized routines... so back to linux, really it is great, it's just not perfect (until the next version, anyway). ;^ )
Sorry, I almost *never* answer the question. You should just ask Matt what headaches linux has caused him, he's got some great stories to tell. ch'! #! * (yes, those are my sound effects) On Tue, 2002-03-12 at 21:53, Larry Price wrote: > Could you elaborate please? What were the erroneous notions you were > entertaining and what information caused you to correct them? -- -- Ben Barrett Software & Systems Engineer counterclaim Phone: 541.484.9235 Fax: 541.484.9193
