[lustre-discuss] Lustre client on kernel.org 4.4 ?
Hi all, As Lustre 2.10 is supported on Ubuntu 16.04 and SLES12 with both based on the 4.4 kernel. I tried to compile the 2.10.6 client using DKMS on the latest Kernel.org 4.4 LTS (4.4-174). But it is failing with what looks like differences in the API (full output attached). make[3]: Entering directory `/usr/src/kernels/4.4.174-1.el6.elrepo.x86_64' LD /var/lib/dkms/lustre-client/2.10.6/build/built-in.o LD /var/lib/dkms/lustre-client/2.10.6/build/libcfs/built-in.o LD /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/built-in.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-tracefile.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-debug.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-prim.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-cpu.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.o cc1: warnings being treated as errors /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.c: In function ‘cfs_access_process_vm’: /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.c:157: error: passing argument 6 of ‘get_user_pages’ makes pointer from integer without a cast include/linux/mm.h:1200: note: expected ‘struct page **’ but argument is of type ‘int’ /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.c:157: error: passing argument 7 of ‘get_user_pages’ from incompatible pointer type include/linux/mm.h:1200: note: expected ‘struct vm_area_struct **’ but argument is of type ‘struct page **’ /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.c:157: error: too many arguments to function ‘get_user_pages’ make[6]: *** [/var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.o] Error 1 make[5]: *** [/var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs] Error 2 make[4]: *** [/var/lib/dkms/lustre-client/2.10.6/build/libcfs] Error 2 make[3]: *** [_module_/var/lib/dkms/lustre-client/2.10.6/build] Error 2 make[3]: Leaving directory `/usr/src/kernels/4.4.174-1.el6.elrepo.x86_64' make[2]: *** [modules] Error 2 make[2]: Leaving directory `/var/lib/dkms/lustre-client/2.10.6/build' make[1]: *** [all-recursive] Error 1 make[1]: Leaving directory `/var/lib/dkms/lustre-client/2.10.6/build' make: *** [all] Error 2 Is there a way to get that to work ? Full disclosure: I need to upgrade the kernel of some Centos6 boxes due to a bug (not Lustre related), but can't upgrade the OS for *reasons*. Best regards, Louis DKMS make.log for lustre-client-2.10.6 for kernel 4.4.174-1.el6.elrepo.x86_64 (x86_64) Mon Feb 11 17:54:21 GMT 2019 make all-recursive make[1]: Entering directory `/var/lib/dkms/lustre-client/2.10.6/build' Making all in . make[2]: Entering directory `/var/lib/dkms/lustre-client/2.10.6/build' make LDFLAGS= CC="gcc" -C /usr/src/kernels/4.4.174-1.el6.elrepo.x86_64 \ -f /var/lib/dkms/lustre-client/2.10.6/build/build/Makefile LUSTRE_LINUX_CONFIG=/usr/src/kernels/4.4.174-1.el6.elrepo.x86_64/.config \ LINUXINCLUDE='-I$(srctree)/arch/$(SRCARCH)/include -Iarch/$(SRCARCH)/include/generated -Iinclude $(if $(KBUILD_SRC),-Iinclude2 -I$(srctree)/include) -I$(srctree)/arch/$(SRCARCH)/include/uapi -Iarch/$(SRCARCH)/include/generated/uapi -I$(srctree)/include/uapi -Iinclude/generated/uapi -include /usr/src/kernels/4.4.174-1.el6.elrepo.x86_64/include/linux/kconfig.h' \ M=/var/lib/dkms/lustre-client/2.10.6/build -o tmp_include_depends -o scripts -o \ include/config/MARKER modules make[3]: Entering directory `/usr/src/kernels/4.4.174-1.el6.elrepo.x86_64' LD /var/lib/dkms/lustre-client/2.10.6/build/built-in.o LD /var/lib/dkms/lustre-client/2.10.6/build/libcfs/built-in.o LD /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/built-in.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-tracefile.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-debug.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-prim.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-cpu.o CC [M] /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.o cc1: warnings being treated as errors /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.c: In function âcfs_access_process_vmâ: /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.c:157: error: passing argument 6 of âget_user_pagesâ makes pointer from integer without a cast include/linux/mm.h:1200: note: expected âstruct page **â but argument is of type âintâ /var/lib/dkms/lustre-client/2.10.6/build/libcfs/libcfs/linux/linux-curproc.c:157: error: passing argument 7 of âget_user_pagesâ from incompatible pointer type include/linux/mm.h:1200: note: expected âstruct vm_area_struct
[lustre-discuss] obdfilter/mdt stats meaning ?
Hi all, I am trying to make sense of some of the OST/MDT stats for 2.12. Can anybody point me to the doc that explain what the metrics are ? The wiki only mention read/write/get_info : http://wiki.lustre.org/Lustre_Monitoring_and_Statistics_Guide But the list I get is quite different : obdfilter.OST001.stats= snapshot_time 1563285450.647120173 secs.nsecs read_bytes340177708 samples [bytes] 4096 4194304 396712660910080 write_bytes 30008856 samples [bytes] 24 4194304 78618271501667 setattr 1755 samples [reqs] punch 73463 samples [reqs] sync 50606 samples [reqs] destroy 31990 samples [reqs] create956 samples [reqs] statfs75378743 samples [reqs] connect 5798 samples [reqs] reconnect 3242 samples [reqs] disconnect5820 samples [reqs] statfs3737980 samples [reqs] preprw370186566 samples [reqs] commitrw 370186557 samples [reqs] ping 882096292 samples [reqs] For the MDT, most are pretty much self explanatory, but I'll still be happy to be pointed to some doc. mdt.MDT.md_stats= snapshot_time 1563287416.006001068 secs.nsecs open 3174644054 samples [reqs] close 3174494603 samples [reqs] mknod 107564 samples [reqs] unlink99625 samples [reqs] mkdir 199643 samples [reqs] rmdir 45021 samples [reqs] rename12728 samples [reqs] getattr 50227431 samples [reqs] setattr 103435 samples [reqs] getxattr 9051470 samples [reqs] setxattr 14 samples [reqs] statfs7525513 samples [reqs] sync 20597 samples [reqs] samedir_rename207 samples [reqs] crossdir_rename 12521 samples [reqs] And anyone knows how to read the OST brw_stats ? obdfilter.OST0014.brw_stats= snapshot_time: 1563287631.511085465 (secs.nsecs) read | write pages per bulk r/w rpcs % cum % | rpcs% cum % 1: 231699298 66 66 | 180944 0 0 2: 855611 0 67 | 322359 1 1 4: 541749 0 67 | 5539716 18 20 8: 1281219 0 67 | 67837 0 20 16: 637808 0 67 | 114546 0 20 32:1342813 0 68 | 3099780 10 31 64:1559834 0 68 | 173166 0 31 128: 1583127 0 69 | 211512 0 32 256: 10627583 3 72 | 499978 1 34 512: 3909601 1 73 | 1029686 3 37 1K: 92141161 26 100 | 18788597 62 100 read | write discontiguous pagesrpcs % cum % | rpcs% cum % 0: 346179839 100 100 | 180946 0 0 1: 0 0 100 | 322363 1 1 2: 0 0 100 | 5521062 18 20 3: 0 0 100 | 18650 0 20 4: 0 0 100 | 18159 0 20 5: 0 0 100 | 26664 0 20 6: 0 0 100 | 10830 0 20 7: 0 0 100 | 12189 0 20 8: 0 0 100 | 11365 0 20 9: 0 0 100 | 10253 0 20 10: 0 0 100 | 8810 0 20 11: 0 0 100 | 9825 0 20 12: 0 0 100 | 16740 0 20 13: 0 0 100 | 14421 0 20 14: 0 0 100 | 10513 0 20 15: 0 0 100 | 32655 0 20 16: 0 0 100 | 1418677 4 25 17: 0 0 100 | 1477077 4 30 18: 0 0 100 | 6227 0 30 19: 0 0 100 | 7071 0 30 20: 0 0 100 | 7297 0 30 21: 0 0 100 | 8478 0 30 22: 0 0 100 | 34591 0 30 23: 0 0 100 | 35591 0 30 24: 0 0 100 | 8378 0 30 25: 0 0 100 | 8724 0 30 26: 0 0 100 | 52300 0 30 27: 0 0 100 | 14038 0 30 28: 0 0 100 | 4734 0 30 29: 0 0 100 | 4878 0 31 30: 0 0 100 | 6232 0 31 31: 0 0 100 | 20708383 68 100 read | write disk I/Os in flightios % cum % | ios % cum % 1: 211177215 61 61 | 29305564 97 97 2:41332944 11 72 | 498260 1 99 3:
Re: [lustre-discuss] [External] Re: obdfilter/mdt stats meaning ?
Hi Aurélien, Thanks for the prompt reply. For the ost stats, any idea what the preprw and commitrw mean ? And why there are two entries with different values for statfs ? For brw_stats even with the doc I still struggle to read this. For example how do you make sense of disk I/O in flight ? read | write disk I/Os in flightios % cum % | ios % cum % 1: 211177215 61 61 | 29305564 97 97 2:41332944 11 72 | 498260 1 99 [..] Does these lines means : Since last snapshot there was 211177215x1 and read 41332944x2 I/O in flight ? Best regards, Louis On 16/07/2019 15:50, Degremont, Aurelien wrote: Hi Louis, About brw_stats, there are a bit of explanation in the Lustre Doc (not that detailed, but still) http://doc.lustre.org/lustre_manual.xhtml#dbdoclet.50438271_55057 > Last thing, is there any way to get the name of the filesystem an OST is part > of by using lctl ? I don't know what you want exactly, but the OST names are self explanatory, there always are like: fsname-OST Where fsname is the lustre filesystem they are part of. For obdfilter stats, these are mostly action to OST objects or client connection management RPCs. setattr: changing an OST object attributes (owner, group, ...) punch: mostly used for truncate (theorically can do holes in files, like truncate with a start and length) sync: straighforward, sync OST to disk destroy: delete an OST object (mostly when a file is deleted) create: create an OST object statfs: like 'df' for this specific OST (used by 'lfs df' by example) (re)connect: when a client connect/reconnect to this OST ping: when a client ping this OST. Aurélien De : lustre-discuss <mailto:lustre-discuss-boun...@lists.lustre.org> au nom de Louis Bailleul <mailto:louis.baill...@pgs.com> Date : mardi 16 juillet 2019 à 16:38 À : lustre-discuss <mailto:lustre-discuss@lists.lustre.org> Objet : [lustre-discuss] obdfilter/mdt stats meaning ? Hi all, I am trying to make sense of some of the OST/MDT stats for 2.12. Can anybody point me to the doc that explain what the metrics are ? The wiki only mention read/write/get_info : http://wiki.lustre.org/Lustre_Monitoring_and_Statistics_Guide<https://urldefense.proofpoint.com/v2/url?u=http-3A__wiki.lustre.org_Lustre-5FMonitoring-5Fand-5FStatistics-5FGuide&d=DwMGaQ&c=KV_I7O14pmwRcmAVyJ1eg4Jwb8Y2JAxuL5YgMGHpjcQ&r=FTXmt89oLXmbXfP78w86-PxB1XdLYgxG8hEoAnZvCvs&m=UC1t7z9tgmxUE2FWaTFHFT_Y69z_VMH0dEYF1VXadX0&s=cdXTUStD_NPwj3GtNYBqJA2nkJ1Ec53F9aD5UxFo5tw&e=> But the list I get is quite different : obdfilter.OST001.stats= snapshot_time 1563285450.647120173 secs.nsecs read_bytes340177708 samples [bytes] 4096 4194304 396712660910080 write_bytes 30008856 samples [bytes] 24 4194304 78618271501667 setattr 1755 samples [reqs] punch 73463 samples [reqs] sync 50606 samples [reqs] destroy 31990 samples [reqs] create956 samples [reqs] statfs75378743 samples [reqs] connect 5798 samples [reqs] reconnect 3242 samples [reqs] disconnect5820 samples [reqs] statfs3737980 samples [reqs] preprw370186566 samples [reqs] commitrw 370186557 samples [reqs] ping 882096292 samples [reqs] For the MDT, most are pretty much self explanatory, but I'll still be happy to be pointed to some doc. mdt.MDT.md_stats= snapshot_time 1563287416.006001068 secs.nsecs open 3174644054 samples [reqs] close 3174494603 samples [reqs] mknod 107564 samples [reqs] unlink99625 samples [reqs] mkdir 199643 samples [reqs] rmdir 45021 samples [reqs] rename12728 samples [reqs] getattr 50227431 samples [reqs] setattr 103435 samples [reqs] getxattr 9051470 samples [reqs] setxattr 14 samples [reqs] statfs7525513 samples [reqs] sync 20597 samples [reqs] samedir_rename207 samples [reqs] crossdir_rename 12521 samples [reqs] And anyone knows how to read the OST brw_stats ? obdfilter.OST0014.brw_stats= snapshot_time: 1563287631.511085465 (secs.nsecs) read | write pages per bulk r/w rpcs % cum % | rpcs% cum % 1: 231699298 66 66 | 180944 0 0 2: 855611 0 67 | 322359 1 1 4: 541749 0 67 | 5539716 18 20 8: 128121
[lustre-discuss] Very bad lnet ethernet read performance
Hi all, I am trying to understand what I am doing wrong here. I have a Lustre 2.12.1 system backed by NVME drives under zfs for which obdfilter-survey gives descent values ost 2 sz 536870912K rsz 1024K obj2 thr 256 write 15267.49 [6580.36, 8664.20] rewrite 15225.24 [6559.05, 8900.54] read 19739.86 [9062.25, 10429.04] But my actual Lustre performances are pretty poor in comparison (can't top 8GB/s write and 13.5GB/s read) So I started to question my lnet tuning but playing with peer_credits and max_rpc_per_pages didn't help. My test setup consist of 133x10G Ethernet clients (uplinks between end devices and OSS are 2x100G for every 20 nodes). The single OSS is fitted with a bonding of 2x100G Ethernet. I have tried to understand the problem using lnet_selftest but I'll need some help/doco as this doesn't make sense to me. Testing a single 10G client [LNet Rates of lfrom] [R] Avg: 2231 RPC/s Min: 2231 RPC/s Max: 2231 RPC/s [W] Avg: 1156 RPC/s Min: 1156 RPC/s Max: 1156 RPC/s [LNet Bandwidth of lfrom] [R] Avg: 1075.16 MiB/s Min: 1075.16 MiB/s Max: 1075.16 MiB/s [W] Avg: 0.18 MiB/s Min: 0.18 MiB/s Max: 0.18 MiB/s [LNet Rates of lto] [R] Avg: 1179 RPC/s Min: 1179 RPC/s Max: 1179 RPC/s [W] Avg: 2254 RPC/s Min: 2254 RPC/s Max: 2254 RPC/s [LNet Bandwidth of lto] [R] Avg: 0.19 MiB/s Min: 0.19 MiB/s Max: 0.19 MiB/s [W] Avg: 1075.17 MiB/s Min: 1075.17 MiB/s Max: 1075.17 MiB/s With 10x10G clients : [LNet Rates of lfrom] [R] Avg: 1416 RPC/s Min: 1102 RPC/s Max: 1642 RPC/s [W] Avg: 708 RPC/s Min: 551 RPC/s Max: 821 RPC/s [LNet Bandwidth of lfrom] [R] Avg: 708.20 MiB/s Min: 550.77 MiB/s Max: 820.96 MiB/s [W] Avg: 0.11 MiB/s Min: 0.08 MiB/s Max: 0.13 MiB/s [LNet Rates of lto] [R] Avg: 7084 RPC/s Min: 7084 RPC/s Max: 7084 RPC/s [W] Avg: 14165RPC/s Min: 14165RPC/s Max: 14165RPC/s [LNet Bandwidth of lto] [R] Avg: 1.08 MiB/s Min: 1.08 MiB/s Max: 1.08 MiB/s [W] Avg: 7081.86 MiB/s Min: 7081.86 MiB/s Max: 7081.86 MiB/s With all 133x10G clients: [LNet Rates of lfrom] [R] Avg: 510 RPC/s Min: 98 RPC/s Max: 23457RPC/s [W] Avg: 510 RPC/s Min: 49 RPC/s Max: 45863RPC/s [LNet Bandwidth of lfrom] [R] Avg: 169.87 MiB/s Min: 48.77MiB/s Max: 341.26 MiB/s [W] Avg: 169.86 MiB/s Min: 0.01 MiB/s Max: 22757.92 MiB/s [LNet Rates of lto] [R] Avg: 23458RPC/s Min: 23458RPC/s Max: 23458RPC/s [W] Avg: 45876RPC/s Min: 45876RPC/s Max: 45876RPC/s [LNet Bandwidth of lto] [R] Avg: 341.12 MiB/s Min: 341.12 MiB/s Max: 341.12 MiB/s [W] Avg: 22758.42 MiB/s Min: 22758.42 MiB/s Max: 22758.42 MiB/s So if I add clients the aggregate write bandwidth somewhat stacks, but the read bandwidth decrease ??? When throwing all the nodes at the system, I am pretty happy with the ~22GB/s on write pretty as this is in the 90% of the 2x100G, but the 341MB/s read sounds very weird considering that this is a third of the performance of a single client. This are my ksocklnd tuning : # for i in /sys/module/ksocklnd/parameters/*; do echo "$i : $(cat $i)"; done /sys/module/ksocklnd/parameters/credits : 1024 /sys/module/ksocklnd/parameters/eager_ack : 0 /sys/module/ksocklnd/parameters/enable_csum : 0 /sys/module/ksocklnd/parameters/enable_irq_affinity : 0 /sys/module/ksocklnd/parameters/inject_csum_error : 0 /sys/module/ksocklnd/parameters/keepalive : 30 /sys/module/ksocklnd/parameters/keepalive_count : 5 /sys/module/ksocklnd/parameters/keepalive_idle : 30 /sys/module/ksocklnd/parameters/keepalive_intvl : 5 /sys/module/ksocklnd/parameters/max_reconnectms : 6 /sys/module/ksocklnd/parameters/min_bulk : 1024 /sys/module/ksocklnd/parameters/min_reconnectms : 1000 /sys/module/ksocklnd/parameters/nagle : 0 /sys/module/ksocklnd/parameters/nconnds : 4 /sys/module/ksocklnd/parameters/nconnds_max : 64 /sys/module/ksocklnd/parameters/nonblk_zcack : 1 /sys/module/ksocklnd/parameters/nscheds : 12 /sys/module/ksocklnd/parameters/peer_buffer_credits : 0 /sys/module/ksocklnd/parameters/peer_credits : 128 /sys/module/ksocklnd/parameters/peer_timeout : 180 /sys/module/ksocklnd/parameters/round_robin : 1 /sys/module/ksocklnd/parameters/rx_buffer_size : 0 /sys/module/ksocklnd/parameters/sock_timeout : 50 /sys/module/ksocklnd/parameters/tx_buffer_size : 0 /sys/module/ksocklnd/parameters/typed_conns : 1 /sys/module/ksocklnd/parameters/zc_min_payload : 16384 /sys/module/ksocklnd/parameters/zc_recv : 0 /sys/module/ksocklnd/parameters/zc_recv_min_nfrags : 16 Best regards, Louis ___ lustre-discuss mailing list lustre-discuss@lists.lustre.org http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
Re: [lustre-discuss] [External] Re: Very bad lnet ethernet read performance
Thanks for the pointers. Flow control has limited impact at this point (no change under lnet_selftest and ~10% drop when disabled under iperf). All machines have tcp_sack enabled. Checksum don't seems to make a difference either. Bumping up the max_rpc_in_flights didn't improve much but seems to have made the write speed more consistent. read_ahead had no effect on read performance. At this point I am struggling to understand what has actual effects on reads. iperf between clients and OSS gives a combined bandwidth that reach ~90% of link capacity (43.7GB/s), but lnet_selftest max out at ~14GB/s so about 28%. Any clues on what lnet tunables / settings could have any impacts here ? Best regards, Louis On 13/08/2019 12:53, Raj wrote: Louis, I would also try: - turning on selective ack (net.ipv4.tcp_sack=1) on all nodes. This helps although there is a CVE out there for older kernels. - turning off checksum osc.ostid*.checksums. This can be turned off per OST/FS on clients. - Increasing max_pages_per_rpc to 16M. Although this may not help with your reads. - Increasing max_rpcs_in_flight and max_dirty_mb be 2 x max_rpcs_in_flight - Increasing llite.ostid*.max_read_ahead_mb to up to 1024 on clients. Again this can be set per OST/FS. _Raj On Mon, Aug 12, 2019 at 12:12 PM Shawn Hall mailto:shawn.h...@nag.com>> wrote: Do you have Ethernet flow control configured on all ports (especially the uplink ports)? We’ve found that flow control is critical when there are mismatched uplink/client port speeds. Shawn From: lustre-discuss mailto:lustre-discuss-boun...@lists.lustre.org>> On Behalf Of Louis Bailleul Sent: Monday, August 12, 2019 1:08 PM To: lustre-discuss@lists.lustre.org<mailto:lustre-discuss@lists.lustre.org> Subject: [lustre-discuss] Very bad lnet ethernet read performance Hi all, I am trying to understand what I am doing wrong here. I have a Lustre 2.12.1 system backed by NVME drives under zfs for which obdfilter-survey gives descent values ost 2 sz 536870912K rsz 1024K obj2 thr 256 write 15267.49 [6580.36, 8664.20] rewrite 15225.24 [6559.05, 8900.54] read 19739.86 [9062.25, 10429.04] But my actual Lustre performances are pretty poor in comparison (can't top 8GB/s write and 13.5GB/s read) So I started to question my lnet tuning but playing with peer_credits and max_rpc_per_pages didn't help. My test setup consist of 133x10G Ethernet clients (uplinks between end devices and OSS are 2x100G for every 20 nodes). The single OSS is fitted with a bonding of 2x100G Ethernet. I have tried to understand the problem using lnet_selftest but I'll need some help/doco as this doesn't make sense to me. Testing a single 10G client [LNet Rates of lfrom] [R] Avg: 2231 RPC/s Min: 2231 RPC/s Max: 2231 RPC/s [W] Avg: 1156 RPC/s Min: 1156 RPC/s Max: 1156 RPC/s [LNet Bandwidth of lfrom] [R] Avg: 1075.16 MiB/s Min: 1075.16 MiB/s Max: 1075.16 MiB/s [W] Avg: 0.18 MiB/s Min: 0.18 MiB/s Max: 0.18 MiB/s [LNet Rates of lto] [R] Avg: 1179 RPC/s Min: 1179 RPC/s Max: 1179 RPC/s [W] Avg: 2254 RPC/s Min: 2254 RPC/s Max: 2254 RPC/s [LNet Bandwidth of lto] [R] Avg: 0.19 MiB/s Min: 0.19 MiB/s Max: 0.19 MiB/s [W] Avg: 1075.17 MiB/s Min: 1075.17 MiB/s Max: 1075.17 MiB/s With 10x10G clients : [LNet Rates of lfrom] [R] Avg: 1416 RPC/s Min: 1102 RPC/s Max: 1642 RPC/s [W] Avg: 708 RPC/s Min: 551 RPC/s Max: 821 RPC/s [LNet Bandwidth of lfrom] [R] Avg: 708.20 MiB/s Min: 550.77 MiB/s Max: 820.96 MiB/s [W] Avg: 0.11 MiB/s Min: 0.08 MiB/s Max: 0.13 MiB/s [LNet Rates of lto] [R] Avg: 7084 RPC/s Min: 7084 RPC/s Max: 7084 RPC/s [W] Avg: 14165RPC/s Min: 14165RPC/s Max: 14165RPC/s [LNet Bandwidth of lto] [R] Avg: 1.08 MiB/s Min: 1.08 MiB/s Max: 1.08 MiB/s [W] Avg: 7081.86 MiB/s Min: 7081.86 MiB/s Max: 7081.86 MiB/s With all 133x10G clients: [LNet Rates of lfrom] [R] Avg: 510 RPC/s Min: 98 RPC/s Max: 23457RPC/s [W] Avg: 510 RPC/s Min: 49 RPC/s Max: 45863RPC/s [LNet Bandwidth of lfrom] [R] Avg: 169.87 MiB/s Min: 48.77MiB/s Max: 341.26 MiB/s [W] Avg: 169.86 MiB/s Min: 0.01 MiB/s Max: 22757.92 MiB/s [LNet Rates of lto] [R] Avg: 23458RPC/s Min: 23458RPC/s Max: 23458RPC/s [W] Avg: 45876RPC/s Min: 45876RPC/s Max: 45876RPC/s [LNet Bandwidth of lto] [R] Avg: 341.12 MiB/s Min: 341.12 MiB/s Max: 341.12 MiB/s [W] Avg: 22758.42 MiB/s Min: 22758.42 MiB/s Max: 22758.42 MiB/s So if I add clients the aggregate write bandwidth somewhat stacks, but the read bandwidth decrease ??? When throwing all the nodes at the system, I am pretty happy with the ~22GB/s on write pretty as this is in the 90% of the 2x100G, but the 341MB/s read sounds very weird considering that this is a third of the performance of a single client. This are my ksocklnd tuning : # for i i