http://fm.ea-tel.eu/flashmedia/flashmeeting/fm19014_1d0116-19014 http://fm.ea-tel.eu/fm/memo.php?pwd=1d0116-19014 http://fm.ea-tel.eu/public The Colorado Front Range OpenSolaris User Group presents a talk by Brendan Gregg. Final edited video with slides can be found at blogs.sun.com slash video brendan gregg dtrace frosug kernel opensolaris performance solaris 0:698:0 766:2046:0 2082:10738:0 1555:10489:0 1760:10739:0 1826:1850:0 1910:10739:0 1930:10738:0 1972:5631:0 2290:2337:0 2989:3487:0 3505:10738:0 5863:5889:0 7968:8035:0 8673:9421:0 9439:9591:0 9607:10739:0 10398:10499:0 lisa are you getting sound okay? Hi. No, unfortunately not. It may be something on my end though. I do hear sound now, but it is very choppy. Deirdre... Sounds good. okay now? I can't hear Brendan very well. I could hear Lynn well. now? About the same (Still not great) I'm hearing him, but he's soudning distant now? Hasn't changed much. sory, can't monitor from my damn laptop, dunno why ..getting better though When he is over by the podium it seems better (for some reason) The room mic. picks him up by the podium? thanks, I know whats' going on will have to drop out and come back, sorry The slides are at http://hub.opensolaris.org/bin/download/User+Group+frosug/meetings/frosug%2D091029.odp Thanks Clay. Wow... Much better sound now! I screwed up a setting, sorry Also, we're on slide 7 (bouncing back to 6 for laughing at SMC more) No prob Deirdre. Sounds great now. slide 13 now I'm getting a lot of static on the line hmm anyone else? No. hmm perhaps its my connection Deirdre could you capture his screen for the demo off the projector? Thank you For the blog which Brendan was talking about see http://blogs.sun.com/brendan/entry/7410_hardware_update_and_analyzing thanks Clay Question was is DTrace better than the previous tools for accuracy whats that command? Kevin Sorry the bash command was basically while [[ 1 == 1 ]]; do :; done & I don't recall the exact comand thanx That was for burning a core. fast Fourier transform would be brutal Arnaud Just pointed out a JVM with large memory could be bad too as the garbage collection time would be large too Arnaud runs into all the items on slide 30 are the slides availabe here ? ..and can we submit questions to Brendan in the Q&A section ? Kevin I'll take questions for Brendan just ask here where are the slides? The slides are available athttp://hub.opensolaris.org/bin/download/User+Group+frosug/meetings/frosug%2D091029.odp thnx The image is at http://blogs.sun.com/brendan/resource/latency-2/latency-2-x2-crop.png The blog being talked about is now http://blogs.sun.com/brendan/entry/latency_art_rainbow_pterodactyl Now talking about the entry http://blogs.sun.com/brendan/entry/unusual_disk_latency Hi Diana! Thank you for joining Video is off the last blog post Any questions? We've moved to the thank you slide i have one Cool, start typing it out and I'll proxy for you. Sadly my laptop speakers stink or we could use your microphone :) ZFS latency with NVRAM arrays Of course this is a little above my head :) Kevin Go ahead Diana ah but you put this together. Thank you question in room: does overclocking help? why would we seeing over 1 ms latency with NVRAM array with 64GB of cache suppose it was memory bound eg graphics program Kevin what type of system are the arrays on and what type of work load? ..and we are only doing write work load (write cache is only 10% used on the array) its a STK 9985V array and its a trading applications, doing about 202 IOPS each being about 12K Kevin, raise your hand to speak and ask the question to the room Kevin we'll try to patch your microphone into our room if you have one i have one hang on go ahead ok KEvin did u get that Do you mean 202k IOPS? We did get it! :D no just 202IOps Very small number of IOps btw, we don't have log device gr8!! thnx Another question in the room is trouble understanding how the zil (ZFS intent log) works when the system doesn't have time to "catch up" and dump the log to disk ? whether the kstat interface is stable this stream will end in ~4 minutes I have a feeling we won't be finished in the room here ...but we'll have video thnx. this streaming was useful