LSE Call on 1/26/01 -------------------- o Attendees: John Hansen, Luann Guleserian, John Wright, Tim Wright, Hans Tannenberg o JH: Gave Update - Steve in Class today. - Steve working on PRD, some effort SWAGs o IBM - Hans - actively hiring College side and internal recruiting is going well Outside - reqs open but not lot of talented people in the market. o JW-SGI under hiring freeze also. Did hire a few people in last few weeks but not necess. just for Scalability JW said he could send some candidates over to us. Everyone agrees. Send over to us. o TW- Update on Scheduler Reaction to posting, latest MQ is full functional equiv. of orig. scheduler. Code posted along with BM results. First posting didn't get a single response. This time, lot of response from mailing list. Lot of interest. Work going on. Regression in performance, rescheduling too often. Some insight from SCO.... Pre-empting investigation. Chat benchmark degradation in performance. Producers need higher priority than consumers.... Running all on same systems, producer and consumer interfering with each other. SchedYield usefulness.....Ingo's patch in 2.4.....removing optimization.... Posting response....difference was publishing benchmark numbers the second time around. Enticed some other people from the community to get involved and give feedback. Lots of good ideas being thrown into the pot. LSE tech mailing list is archived......Nakajima, Andy Kleem, Marcello - good stuff on VM side. Rick Van Riel involved in XFS stuff. Very good ally. Some sites out there that do threaded kernel archives online. Mailing list archive is off of Project Resources on the main web page. o JW on Scheduler work. John Hawkes trying to apply the patch on mips64 and get a benchmark run. Some troubles with the machien and lost it for a week. Numbers coming next week on higher processor count. o TW - IA64 stuff, Hans: able to put a pre-2.4 kernel on an 8processor numa ia64 and working on a 12p ia64 numa. IA64 work is just a paralel effort in kernel. Not really intended for Scalability. Just in bring up mode. o JW - SGI still a ways from IA64 - having numbers. o TW - IA64 will become more significant. o JW - SGI will be working on IA64. o TW - IA64 Linux will be a serious platform. o TW - Jonathan working on the 4way with his db work. Going from 2way to 4way Interesting was scheduler ....scheduler....kiobuf patches. o JW - KIOBUF work progressing but not much use of it yet. o TW - Locking in ext2 is wrong. o JW - KIOBUF....Christenson (Caldera) and Suse interested in cleaning things up...handing off maintenance and putting up on SourceForge site. Slated to go in as soon as 2.5 opens. In XFS, huge boost in performance on that. o TW - Rick not fully engaged but quite a ways on documenting the locks and locking hier. within the kernel. 1/3 through and accelerating. Hoping to have a draft to review in March unless he gets yanked. JW put a placeholder on the web for that. o JW - Paul Mckenney? copylock....guys helping him...started the port...on the way...code will be ready for people to play with and intent to release under gpl. Trying to get lawyers to approve due to copyrights and patents. o TW - Other Areas - not directly Scalability Hot Plug PCI - lots of machines and no support. On the horizon. o JW - Platform team AR-JohnH to come back with an update on Hot Plug PCI to this forum. Intel's Actities. o LG - Compaq's 8way different than Intel's, don't take Intel Motherboards. o TW - 2.4 System Management Issues ServerWorks chipset issues Machines with SystMgmt built into motherboard. works fine with 2.2. o JW - Is Dave Olien? TW: working on Interrupt, APIC, PIC, haven't synched with lately Looking at pinging interrupts? Nick Pollett sent out info on process pinging.. TW: ......interrupts get steered to cpu...(hack)...dont like it but workload is web servers (popular - big win)....feeling now is it won't be close enough to satisfy people due to cache issues. Kind of ugly. 4 1ways with shared memory. o JW - kind of like numa stuff... want to make it flexible to extend and not have to rewrite. Will follow up with Nick, maybe some ideas. Didn't see any postings from Dave. Maybe Nick can contact him directly. o Hans - what are SGI doing on numa API? Any looking at it? Suse also pushing for it? o JW - Scott Fainer looking at few issues....Kanoj is the other, but in India trying to get Green card. Will be back. Scott looking at scheduling calls into an arch. specific alg. but how does it work into multiq sched. work? Any ideas? no... Philosophy: make the simplest change that will get accepted... o TW - saw that and made sense. o JW: John hawkes talking about pluggable algorithms....replaceable files to allow specific schedulers for diff. arch.s o TW - HP going to demo pluggable scheduler next week at LW. o JW: might need a pow-wow. o JH: Anyone at LW? Tim and Hans and Hubertus. AR - John H. to send out info to Hans and Tim on the Itanium BOF and OSDL BOF. o JW - Another confernce is Ottowa Linux symposium - no trade show floor. Very informal and technical. Really geared for engineers and many good people there. Web is www.ottawalinuxsymposium.org o LG - we need to send some people. AR-John and Luann to get the word out at Intel on Ottowa Linux Symposium. o Hans - at least 5 people signed up to attend OLS at IBM. o JW - pinging around ideas on numa API, Irix has a .... mechanism and different but similiar to what everyone else does. Open to ideas, will have first work up there soon. Want it flexible. It won't be the irix version......May have a specific discussion for this. Watch the mailing list. Maybe 1.5 hrs. Sounds good for IBM folks. Conf. call for techies. Probably useful for people to setup BOFs for some of these projects. o JW - action item to do X SCSI....Justin Gibbs at Adaptec and Eric Youngdale. Maintainer....but not engaged on rewriting efforts. Arch/reviewer. Justin Gibbs has volunteered to rewrite the scsci layer and few prelim ideas. Want to engage and really rewrite it the right way. o TW - Justin published BSD adaptec driver.....problems with some drives in the 2.4 kernel....Justin's fixes it.... o JW - right guys to engage....Tim going to release SSI docs? no, no further action.... Hearing very good performance out of x-scsi guys with kio...very good gains...encouraging. o JW - Ulf Carlson working on some large page stuff....Irix does everything, ...swiss army knife of getting VM pages....takes lot of infrast. and questionable ROI but we will pick a few large page sizes and do those. Stay tuned. Prelim code in next week or tw. Gearing it to IA64. Will continue on that path and see where we get. o TW - ia32 is limited....IA64 is more flexible. o JW - Ananth looking at VM issues with Rick and some falling out of XFS project...that going well but not necessarily scaling or numa-wise. o Hans - Kanoj working with Numa api...can you make sure he talks to Hans when he returns? o JW - expects to be in end of month. Scott Fainer looking at it. Will ask him to spit out an update to the list (Kanoj) when he gets back. o Hans - already have Randall Craig...want to make sure more companies joining in to help out and define Numa api's in near future. Randall is from suse. o JW - send info on any people they should also talk to..... o Hans - trying to build up the team, little bit of waiting time, but 2 people coming FT in March and part-time earlier. Names not finalized. Maybe final in 2 weeks. o JW - will throw out an update when Kanoj gets back and will give the direction and look for feedback. o Hans - Hans or Tim are the right contacts at IBM till the right people identified. o JW - LBS - put some stuff out there, some feedback, no time to update it... o TW - Ray Bryant said whats the charter, why different than 3-4 other ones out there, We think there is a difference, repository for benchmarks for Scalab. Maybe need to make that more clear. o JW - other BM sites geared to eval. PC hardware....want to take bunch of BM's not relevant to our work and push off to "other BM page". Goal is an exhaustive list of BM's, recommended list of open BMs and how to run them with scripts to collect data. Has the email from Ray and will respond. Knew it was sloppy but trying to get something out there. Will get back to it next week.