Back to Home / #uml / 2007 / 04 / Prev Day | Next Day
#uml IRC Logs for 2007-04-20

---Logopened Fri Apr 20 00:00:19 2007
02:51|-|fo0bar [] has quit [Quit: leaving]
02:54|-|fo0bar [] has joined #uml
03:25|-|wyvern [] has quit [Ping timeout: 480 seconds]
03:29|-|wyvern [] has joined #uml
03:46|-|motp [~motp@] has joined #uml
04:09|-|polyonymous [] has quit [Ping timeout: 480 seconds]
04:20|-|polyonymous [] has joined #uml
05:22|-|da-x [] has quit [Ping timeout: 480 seconds]
05:23|-|Ancalagon [] has joined #uml
05:30|-|da-x [] has joined #uml
05:44|-|aroscha [] has quit [Quit: aroscha]
06:06|-|aroscha [] has joined #uml
06:41|-|Ancalagon [] has quit [Quit: ChatZilla [Firefox]]
06:59|-|krau [~cktakahas@] has joined #uml
07:45|-|motp [~motp@] has quit [Quit: Leaving]
08:51|-|jdike [] has joined #uml
08:51<jdike>Hi guys
08:52<caker>hello Jeff
09:02<jdike>I can't see that per-device I/O threads or readv/writev make any difference
09:05<caker>in theory or did you implement it?
09:06<jdike>I implemented it
09:06<jdike>in theory, they should be great
09:07<caker>Even getting funneled down to one device on the host? I could see it having more of an impact (over the single IO thread) if the ubd devices lived on different devices on the host
09:07<jdike>the workloads were - kernel build, dd from device to /dev/null, dd from device to file on same device, dd from one device to another
09:08<caker>yeah -- the dd from one device to another scenario
09:08<jdike>everything is going through the page cache on the host
09:08<jdike>these were ubd devices
09:08<jdike>everything on the same host disk
09:08<caker>I thought the O_DIRECT or AIO took care of the page cache thing
09:08<jdike>no AIO involved here
09:09<jdike>I have the same problem with it
09:09<caker>is this to device nodes on the host or image files?
09:09<jdike>I can't find a performance improvement
09:09<jdike>image files
09:10<caker>maybe try with LVM volumes to elimiate the host page cache?
09:10|-|aroscha [] has quit [Ping timeout: 480 seconds]
09:10<jdike>you're going through the page cache unless you're doing O_DIRECT
09:11<jdike>besides, I want to see a performance benefit without having to trick out the host
09:11<caker>is that accurate? For some reason I thought using dev nodes eliminated the double caching thing
09:11<jdike>no, it doesn't
09:12<jdike>however, device nodes aren't coherent with files on the device
09:12<caker>I guess I'm thinking it gets rid of going through the host VFS
09:13<caker>*no* improvement at all?
09:13<jdike>nothing that jumps off the screen
09:13<caker>worth comparing your tests with ubd devices on separate host devices?
09:13|-|ram [] has joined #uml
09:14<jdike>maybe if I averaged the performance and looked at standard deviations, there might be something
09:14<caker>I guess that's an unlikely situation for most
09:14<jdike>but with these sorts of things, I'm not that interested in 5%
09:14|-|nessie [] has quit [Remote host closed the connection]
09:14|-|nessie [] has joined #uml
09:18|-|hfb [] has quit [Quit: Leaving]
09:34|-|aroscha [] has joined #uml
09:35|-|baroni [] has quit [Remote host closed the connection]
09:36|-|baroni [] has joined #uml
09:37|-|aroscha [] has quit []
10:03|-|hfb [] has joined #uml
10:28|-|aroscha [] has joined #uml
10:33|-|hfb [] has left #uml [Leaving]
11:14|-|richardw [] has joined #uml
11:19|-|Urgleflogue [] has quit [Quit: 01001110 01100101 01110010 01100100 00100001]
11:23|-|Urgleflogue [] has joined #uml
11:25|-|aroscha [] has quit [Quit: aroscha]
11:28|-|richardw [] has quit [Read error: Connection reset by peer]
11:46|-|ram [] has quit [Read error: Operation timed out]
11:46|-|richardw [] has joined #uml
11:52|-|mjf [] has joined #uml
11:54|-|aroscha [] has joined #uml
11:58|-|the_hydra [~a_mulyadi@] has joined #uml
12:20|-|ram [] has joined #uml
12:23|-|amitg [] has quit [Quit: Changing server...]
12:27|-|the_hydra [~a_mulyadi@] has quit [Read error: Connection reset by peer]
12:27|-|the_hydra [~a_mulyadi@] has joined #uml
12:27<the_hydra>baroni: sorry dc
12:31<baroni>the_hydra: np :)
14:04|-|aroscha [] has quit [Quit: aroscha]
14:07|-|aroscha [] has joined #uml
14:19|-|krau [~cktakahas@] has quit [Quit: Leaving]
14:20|-|aroscha [] has quit [Quit: aroscha]
14:33|-|the_hydra [~a_mulyadi@] has quit []
15:04|-|kos_tom [] has joined #uml
15:28|-|aroscha [~aroscha@] has joined #uml
15:34|-|baroni [] has quit [Remote host closed the connection]
15:34|-|baroni [] has joined #uml
16:13<aroscha>jdike: 2000 are running
16:14<aroscha>what do you people actually think of VNUML? Is it usuable? Anybody tried it?
16:15<jdike>looked at vnuml a long time ago, looked neat, but never used it
16:15<aroscha>looks like not completed for me.
16:23|-|jdike [] has quit [Quit: Leaving]
17:26|-|aroscha [~aroscha@] has quit [Ping timeout: 480 seconds]
17:29|-|richardw [] has quit [Quit: Leaving]
17:47|-|kos_tom [] has quit [Quit: I like core dumps]
18:01|-|krau [~cktakahas@] has joined #uml
18:40|-|kos_tom [] has joined #uml
18:43|-|mjf [] has quit [Remote host closed the connection]
18:52|-|kos_tom [] has quit [Quit: I like core dumps]
19:54|-|ram [] has quit [Ping timeout: 480 seconds]
20:45|-|aroscha [] has joined #uml
21:19|-|aroscha [] has quit [Quit: aroscha]
22:59|-|VS_ChanLog [] has left #uml [Rotating Logs]
22:59|-|VS_ChanLog [] has joined #uml
23:50|-|hfb [] has joined #uml
---Logclosed Sat Apr 21 00:00:57 2007