[00:58:43] *** Joins: tomzawadzki (uid327004@gateway/web/irccloud.com/x-fpqnvifzcioelisw) [01:50:18] Are there any sizing guidelines -- that can help you figure out how many huge pages you need for lets say; 1000 LUs iscsi, nvmf ? [01:55:02] gila: no, but it should be fairly easy to check manually [01:55:43] any pointers? [01:55:45] you could setup 2mb hugepages and start your spdk app with -s 0 [01:56:09] right, and. then add lI [01:56:13] right, and. then add LU? [01:56:18] then add a single lun/iscsi target/nvmf target to preallocate any resources [01:56:46] check the amount of mapped hugepages (e.g. with pmap, but there might be better alternatives) [01:57:51] then allocate a couple more luns/iscsi targets/nvmf targets and check pmap again [01:59:18] Once IO starts to flow, the consumed amount of memory does not grow? i.e everything is preallocated? [01:59:55] yup, everything should be preallocated [02:00:47] especially since hugepage allocations are very costly - way more costly than posix malloc [02:04:10] Yeah. Speaking of which. there was this talk about DPDK where they reevaluated the costs of 4kb pages and DPDK and it turned out to be not all that bad for particular workloads, have you seen it? @darsto [02:12:13] gila: I haven't seen that talk, but it sounds interesting [02:12:45] https://dpdksummitnorthamerica2018.sched.com/event/Ihgw/revise-4k-pages-performance-impact-for-dpdk-applications-lei-yao-jiayu-hu-intel [02:13:02] It might be usable (dunno ofcourse) for lets say all iSCSI environments. [02:13:12] I think a video is up somewhere as well. [03:29:02] *** Quits: tomzawadzki (uid327004@gateway/web/irccloud.com/x-fpqnvifzcioelisw) (Quit: Connection closed for inactivity) [03:30:47] *** Joins: tomzawadzki (uid327004@gateway/web/irccloud.com/x-wdwaypxisryfywii) [09:01:00] *** Joins: bwalker_ (~bwalker@192.55.54.38) [09:01:00] *** ChanServ sets mode: +o bwalker_ [09:11:20] I'm going to reboot znc here in a second for those of you on the intel bouncer [09:11:23] it will come right bakc up [12:35:57] *** Joins: bwalker (~bwalker@134.134.139.72) [12:35:57] *** Server sets mode: +cnrt [12:35:57] *** Server sets mode: +cnrt [12:35:57] *** ChanServ sets mode: +o bwalker [12:36:14] *** Quits: bwalker__ (bwalker@nat/intel/x-syvirptemahrvqkk) (Quit: Leaving) [12:41:03] *** Joins: bwalker (~bwalker@134.134.139.72) [12:41:03] *** ChanServ sets mode: +o bwalker [12:41:04] *** Server sets mode: +cnrt [12:41:04] *** Server sets mode: +cnrt [12:52:10] *** Joins: peluse (peluse@nat/intel/x-ncacnempbkzewodg) [12:52:11] *** ChanServ sets mode: +o peluse [12:52:20] bwalker, thanks [12:54:36] *** Joins: ppelplin (~ppelplin@134.134.139.72) [13:31:04] *** Joins: bwalker (~bwalker@134.134.139.72) [13:31:05] *** ChanServ sets mode: +o bwalker [13:31:05] *** Server sets mode: +cnrt [13:31:05] *** Server sets mode: +cnrt [13:37:50] *** Joins: bwalker (~bwalker@134.134.139.72) [13:37:51] *** ChanServ sets mode: +o bwalker [13:37:51] *** Server sets mode: +cnrt [13:37:51] *** Server sets mode: +cnrt [13:42:59] Happy new years folks. Any (imminent?) plans to bring DPDK submodule up to the LTS version (18.11)? [13:47:22] *** Joins: sethhowe (~sethhowe@134.134.139.72) [13:53:26] *** Joins: ppelplin (ppelplin@nat/intel/x-jkcbzwfsyijonrzr) [13:56:48] *** Joins: peluse (peluse@nat/intel/x-fmrqgwdyntnrvjgs) [13:56:48] *** ChanServ sets mode: +o peluse [15:26:11] *** Quits: bluebird (~bluebird@p5B0ACD29.dip0.t-ipconnect.de) (Quit: Leaving) [17:56:10] *** Joins: travis-ci (~travis-ci@ec2-107-23-200-255.compute-1.amazonaws.com) [17:56:11] (spdk/master) test/vm_setup.sh: Add sshpass (Karol Latecki) [17:56:12] Diff URL: https://github.com/spdk/spdk/compare/8aaec569a0cb...2a9667066aab [17:56:12] *** Parts: travis-ci (~travis-ci@ec2-107-23-200-255.compute-1.amazonaws.com) ()