----- Original Message ----- > From: "Atsushi Kumagai" <ats-kumagai at wm.jp.nec.com> > To: "HATAYAMA Daisuke (d.hatayama at jp.fujitsu.com)" <d.hatayama at jp.fujitsu.com>, "Chao Fan" <cfan at redhat.com> > Cc: zhouwj-fnst at cn.fujitsu.com, kexec at lists.infradead.org > Sent: Thursday, December 24, 2015 4:20:42 PM > Subject: RE: [PATCH RFC 00/11] makedumpfile: parallel processing > > >> >> >> >> Could you provide the information of your cpu ? > >> >> >> >> I will do some further investigation later. > >> >> >> >> > >> >> >> > > >> >> >> > OK, of course, here is the information of cpu: > >> >> >> > > >> >> >> > # lscpu > >> >> >> > Architecture: x86_64 > >> >> >> > CPU op-mode(s): 32-bit, 64-bit > >> >> >> > Byte Order: Little Endian > >> >> >> > CPU(s): 48 > >> >> >> > On-line CPU(s) list: 0-47 > >> >> >> > Thread(s) per core: 1 > >> >> >> > Core(s) per socket: 6 > >> >> >> > Socket(s): 8 > >> >> >> > NUMA node(s): 8 > >> >> >> > Vendor ID: AuthenticAMD > >> >> >> > CPU family: 16 > >> >> >> > Model: 8 > >> >> >> > Model name: Six-Core AMD Opteron(tm) Processor 8439 SE > >> >> >> > Stepping: 0 > >> >> >> > CPU MHz: 2793.040 > >> >> >> > BogoMIPS: 5586.22 > >> >> >> > Virtualization: AMD-V > >> >> >> > L1d cache: 64K > >> >> >> > L1i cache: 64K > >> >> >> > L2 cache: 512K > >> >> >> > L3 cache: 5118K > >> >> >> > NUMA node0 CPU(s): 0,8,16,24,32,40 > >> >> >> > NUMA node1 CPU(s): 1,9,17,25,33,41 > >> >> >> > NUMA node2 CPU(s): 2,10,18,26,34,42 > >> >> >> > NUMA node3 CPU(s): 3,11,19,27,35,43 > >> >> >> > NUMA node4 CPU(s): 4,12,20,28,36,44 > >> >> >> > NUMA node5 CPU(s): 5,13,21,29,37,45 > >> >> >> > NUMA node6 CPU(s): 6,14,22,30,38,46 > >> >> >> > NUMA node7 CPU(s): 7,15,23,31,39,47 > >> >> >> > >> >> >> This CPU assignment on NUMA nodes looks interesting. Is it possible > >> >> >> that this affects performance of makedumpfile? This is just a guess. > >> >> >> > >> >> >> Could you check whether the performance gets imporoved if you run > >> >> >> each > >> >> >> thread on the same NUMA node? For example: > >> >> >> > >> >> >> # taskset -c 0,8,16,24 makedumpfile --num-threads 4 -c -d 0 vmcore > >> >> >> vmcore-cd0 > >> >> >> > >> >> > Hi HATAYAMA, > >> >> > > >> >> > I think your guess is right, but maybe your command has a little > >> >> > problem. > >> >> > > >> >> > From my test, the NUMA did affect the performance, but not too much. > >> >> > The average time of cpus in the same NUMA node: > >> >> > # taskset -c 0,8,16,24,32 makedumpfile --num-threads 4 -c -d 0 vmcore > >> >> > vmcore-cd0 > >> >> > is 314s > >> >> > The average time of cpus in different NUMA node: > >> >> > # taskset -c 2,3,5,6,7 makedumpfile --num-threads 4 -c -d 0 vmcore > >> >> > vmcore-cd0 > >> >> > is 354s > >> >> > > >> >> > >> >> Hmm, according to some previous discussion, what we should see here is > >> >> whether it affects performance of makedumpfile with --num-threads 1 > >> >> and -d 31. So you should need to compare: > >> >> > >> >> # taskset 0,8 makedumpfile --num-threads 1 -c -d 31 vmcore > >> >> vmcore-d31 > >> >> > >> >> with: > >> >> > >> >> # taskset 0 makedumpfile -c -d 0 vmcore vmcore-d31 > >> > >> I removed -c option wrongly. What I wanted to write is: > >> > >> # taskset -c 0,8 makedumpfile --num-threads 1 -d 31 vmcore vmcore-d31 > >> > >> and: > >> > >> # taskset -c 0 makedumpfile -d 31 vmcore vmcore-d31 > >> > >> just in case... > > Why did you remove -c option from makedumpfile ? > We are discussing the performance with compression. > I think the below is correct: > > # taskset -c 0,8 makedumpfile --num-threads 1 [-c|-l|-p] -d 31 vmcore > vmcore-d31 > > and: > > # taskset -c 0 makedumpfile [-c|-l|-p] -d 31 vmcore vmcore-d31 > Hi Atsushi Kumagai, "taskset -c 0,8 makedumpfile --num-threads 1" "taskset -c 0 makedumpfile" -c 52s 61s -l 33s 17s -p 33s 18s Thanks, Chao Fan > > Thanks, > Atsushi Kumagai > > >Hi HATAYAMA, > > > >the average time of > ># taskset -c 0,8 makedumpfile --num-threads 1 -d 31 vmcore vmcore-d31 > >is 33s. > >the average time of > ># taskset -c 0 makedumpfile -d 31 vmcore vmcore-d31 > >is 18s. > > > >My test steps: > >1. change /etc/kdump/conf with > >"core_collector makedumpfile -l --message-level 1 -d 31" > >2. make a crash > >3. cd into the directory of the vmcore made by kdump > >4. in the directory of vmcore do > ># taskset -c 0,8 makedumpfile --num-threads 1 -d 31 vmcore vmcore-d31 > >or > ># taskset -c 0 makedumpfile -d 31 vmcore vmcore-d31 > > > >if there are there any problems, please tell me. > > > >Thanks, > >Chao Fan > > > >> >> > >> >> Also, I'm assuming that you've done these benchmark on kdump 1st > >> >> kernel, not kdump 2nd kernel. Is this correct? > >> >> > >> > Hi HATAYAMA, > >> > > >> > I test in the first kernel, not in the kdump second kernel. > >> > > >> > >> I see. > >> > >> -- > >> Thanks. > >> HATAYAMA, Daisuke > >> _______________________________________________ > >> kexec mailing list > >> kexec at lists.infradead.org > >> http://lists.infradead.org/mailman/listinfo/kexec > >> > > > >_______________________________________________ > >kexec mailing list > >kexec at lists.infradead.org > >http://lists.infradead.org/mailman/listinfo/kexec > _______________________________________________ > kexec mailing list > kexec at lists.infradead.org > http://lists.infradead.org/mailman/listinfo/kexec >
