[00:11:21] --- abo has left [00:37:27] --- kaj has become available [01:25:05] --- dev-zero@jabber.org has become available [04:54:33] --- abo has become available [05:19:29] --- kaj has left [05:19:31] --- kaj has become available [05:49:03] --- abo has left [05:49:19] --- abo has become available [06:07:13] --- Simon Wilkinson has become available [06:07:15] --- Simon Wilkinson has left [06:07:20] --- Simon Wilkinson has become available [06:08:38] --- Simon Wilkinson has left [06:45:20] --- abo has left [06:45:33] --- abo has become available [07:22:07] --- mho has become available [07:23:48] --- deason has become available [07:32:33] --- dev-zero@jabber.org has left [08:17:19] do we have an idea of if there will be a pre4 for 1.4.12? and/or when? [08:18:31] I was wondering if it's possible one of the things for 126454 to go in, since it's another hostList corruption issue [08:18:47] "if it's possible for one of the things" [08:29:58] --- dev-zero@jabber.org has become available [08:42:14] --- reuteras has left [08:46:40] --- dev-zero@jabber.org has left [08:52:01] --- dev-zero@jabber.org has become available [09:25:05] --- kaj has left [10:01:04] --- kevin.sumner has become available [10:05:14] --- Russ has become available [10:14:30] --- Russ has left: Disconnected [10:14:40] --- Russ has become available [10:32:55] --- kaj has become available [10:56:24] --- kaj has left [11:02:55] Wow, our fileserver tuning sucks. [11:05:18] - The default number of threads (-p) should be 128, not 6. - The default number of volume attach threads should be more than one, unless we think that parallel attach doesn't work. Of course, DAFS makes this a bit of a moot point. - The default number of directory buffers (-b) should be at least as large as the number of threads; otherwise you risk starvation which, if it happens, will crash the server. [11:06:17] --- kaj has become available [11:06:31] --- kaj has left [11:07:15] - The buffer we use to read and write data in RXAFS_{Fetch,Store}Data, -sendsize, defaults to only 16K! Some of the same arguments for larger chunk size apply here as well, but IMHO with fewer negatives. - We set a UDP packet receive buffer size (-udpsize) of 64K, which is supposed to be "large" but in fact is _smaller_ than the default on most platforms. [11:08:41] - The directory buffering code implements "LRU" by a linear search of _all_ buffers to identify the oldest one. - If you set -udpsize larger than the kernel-imposed maximum, the setsockopt call will fail, and the fileserver will then do another call to set the buffer size to 32K, which is smaller than both our default and most OS's. [11:18:51] --- sxw mobile has become available [11:25:47] --- Russ has left: Disconnected [11:26:22] --- Russ has become available [11:29:28] --- kaj has become available [11:29:35] --- kaj has left [11:32:22] send a patch. or remind me and i will try to do it myself. [11:32:40] you of all people though should not advocate a default change for num threads in a stable version [11:34:26] I'm not advocating a default change to any of this stuff in 1.4.x. But if you know of some reason why -p 128 -vattachpar 8 is not a good idea for a 1.4.x server, say so [11:35:56] --- sxw mobile has left [11:36:04] pre4: still waiting for pre3 review comments. i assume one of the hostList race handling things, the fix for the pruclient thing, and i don't know of anything else. [11:36:18] no, do it [11:36:28] works for me. [11:39:07] I wonder how long we have to wait after the last prerelease to avoid releasing 1.4.14 within a month of 1.4.13. [11:39:26] Er... s/14...13/13...12/ [11:41:07] --- sxw mobile has become available [11:41:14] In case you haven't figured out, I'm doing some fileserver tuning that I should have done long ago. When I'm done, I'll end up with a writeup that explains my rationale, as I did for client tuning some years ago. I'll probably ask people here to review, then post to -info [11:42:32] --- haba has become available [11:50:25] ok. it should actually go in the wiki [11:51:14] Oh, maybe. After discussion. [11:58:42] --- sxw mobile has left [12:03:27] --- sxw mobile has become available [12:04:25] --- sxw mobile has left [12:06:16] --- Russ has left: Disconnected [12:06:52] --- dev-zero@jabber.org has left: Lost connection [12:12:11] --- sxw mobile has become available [12:21:16] --- dev-zero@jabber.org has become available [12:27:28] --- sxw mobile has left [12:39:45] --- sxw mobile has become available [12:41:10] --- dev-zero@jabber.org has left [12:47:59] --- meffie has become available [12:59:11] --- Simon Wilkinson has become available [13:07:49] --- Russ has become available [13:31:39] Is grand's qrunner having trouble again? [13:36:39] --- bpoliakoff has become available [13:37:15] --- bpoliakoff has left [13:57:03] --- dev-zero@jabber.org has become available [13:57:43] --- dev-zero@jabber.org has left [14:03:59] --- dev-zero@jabber.org has become available [14:28:11] --- sxw mobile has left [14:40:51] --- kevin.sumner has left [15:57:12] --- deason has left [17:46:18] --- meffie has left [20:21:21] --- deason has become available [22:23:48] --- reuteras has become available [22:51:30] --- deason has left [23:32:20] --- Simon Wilkinson has left: Lost connection [23:54:54] --- haba has left