So... my first question is still actual. We plan to study source code, but before, maybe. somebody who already know it can tell me - how we can reduce tgtd memory usage? Thank you! 25.03.2011, в 0:10, Stepan Fedorov написал(а): > > OK! I do some tests on my test environment and here is the results: > > My target node has 12Gb of memory. I created 300 LUN's on it - tgtd process virtual size become something above 13Gb. > > Next, i attach this LUNs on three initiator nodes - 100 LUNs per node. > > Now - my task is to initiate some IO on all of this LUNs in parallel, to see - what happen with tgtd. > > To do this, i created LVM striped volume (/dev/vbd0/megadisk0) from 100 iSER-attached LUNs on each initiator node. > > Next, i run e2fsck /dev/vbd0/megadisk0 on three nodes. During this process, i got messages about IO errors on all LUNs in dmesg on all three nodes, because tgtd segfaults on target node, with this message in dmesg: > [734121.705847] tgtd[18959]: segfault at 0 ip (null) sp 00007fff04d22118 error 14 in tgtd[400000+4a000] > > > 24.03.2011, в 14:06, FUJITA Tomonori написал(а): > >> On Thu, 24 Mar 2011 14:09:14 +0300 >> Степан Фёдоров <stepan.fedorov@xxxxxxxx> wrote: >> >>> Hm... i create 666 targets with one LUN per target on node with 12Gb >>> of memory, and in top i see this: >>> >>> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND >>> >>> 7862 >>> root 20 0 22.4g 1.6g 1144 S 2 13.4 7:39.18 >>> /usr/sbin/tgtd >>> >>> VIRT, and RES values grows while i attach this targets on initiator >>> node. What will happen, if tgt begin usage of all this memory while >>> serving it's targets? It will be killed by oom killer? >> >> tgtd can avoid oom killer by setting /proc/{pid}/oom_adj. >> >> >>> Ooops. While i write this - on initiator side i got in dmesg a "BUG" message ) >> >> Needs to report it to open-iscsi mailing list. > > Stepan Fedorov > sf@xxxxxxxx > > > Stepan Fedorov sf@xxxxxxxx -- To unsubscribe from this list: send the line "unsubscribe stgt" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html