<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<div class="moz-cite-prefix">Thanks for the pointers.<br>
<br>
Flow control has limited impact at this point (no change under lnet_selftest and ~10% drop when disabled under iperf).<br>
All machines have tcp_sack enabled.<br>
Checksum don't seems to make a difference either.<br>
Bumping up the max_rpc_in_flights didn't improve much but seems to have made the write speed more consistent.<br>
read_ahead had no effect on read performance.<br>
<br>
At this point I am struggling to understand what has actual effects on reads.<br>
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%.<br>
<br>
Any clues on what lnet tunables / settings could have any impacts here ?<br>
<br>
Best regards,<br>
Louis<br>
<br>
On 13/08/2019 12:53, Raj wrote:<br>
</div>
<blockquote type="cite" cite="mid:CANF66k_qtwzwTKxLSqMA9HuH7kDifJidXWeFpgAi=0spO8LPcw@mail.gmail.com">
<div dir="ltr">
<div dir="ltr">Louis,
<div>I would also try:</div>
<div>- turning on selective ack (net.ipv4.tcp_sack=1) on all nodes. This helps although there is a CVE out there for older kernels.</div>
<div>- turning off checksum osc.ostid*.checksums. This can be turned off per OST/FS on clients.</div>
<div>- Increasing max_pages_per_rpc to 16M. Although this may not help with your reads.</div>
<div>- Increasing max_rpcs_in_flight and max_dirty_mb be 2 x max_rpcs_in_flight</div>
<div>- Increasing llite.ostid*.max_read_ahead_mb to up to 1024 on clients. Again this can be set per OST/FS.</div>
<div><br>
</div>
<div>_Raj</div>
</div>
</div>
<br>
<div class="gmail_quote">
<div dir="ltr" class="gmail_attr">On Mon, Aug 12, 2019 at 12:12 PM Shawn Hall <<a href="mailto:shawn.hall@nag.com" moz-do-not-send="true">shawn.hall@nag.com</a>> wrote:<br>
</div>
<blockquote class="gmail_quote" style="margin:0px 0px 0px
0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex">
<div bgcolor="white" lang="EN-US">
<div class="gmail-m_-346298009108846501WordSection1">
<p class="MsoNormal"><span style="color:windowtext">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.</span></p>
<p class="MsoNormal"><span style="color:windowtext"> </span></p>
<p class="MsoNormal"><span style="color:windowtext">Shawn</span></p>
<p class="MsoNormal"><span style="color:windowtext"> </span></p>
<div>
<div style="border-style:solid none
none;border-top-width:1pt;border-top-color:rgb(225,225,225);padding:3pt
0in 0in">
<p class="MsoNormal"><b><span style="color:windowtext">From:</span></b><span style="color:windowtext"> lustre-discuss <<a href="mailto:lustre-discuss-bounces@lists.lustre.org" target="_blank" moz-do-not-send="true">lustre-discuss-bounces@lists.lustre.org</a>>
<b>On Behalf Of </b>Louis Bailleul<br>
<b>Sent:</b> Monday, August 12, 2019 1:08 PM<br>
<b>To:</b> <a href="mailto:lustre-discuss@lists.lustre.org" target="_blank" moz-do-not-send="true">
lustre-discuss@lists.lustre.org</a><br>
<b>Subject:</b> [lustre-discuss] Very bad lnet ethernet read performance</span></p>
</div>
</div>
<p class="MsoNormal"> </p>
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">Hi all,<br>
<br>
I am trying to understand what I am doing wrong here.<br>
I have a Lustre 2.12.1 system backed by NVME drives under zfs for which obdfilter-survey gives descent values</span></p>
<blockquote style="margin-top:5pt;margin-bottom:5pt">
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">ost 2 sz 536870912K rsz 1024K obj 2 thr 256 write 15267.49 [6580.36, 8664.20] rewrite 15225.24 [6559.05, 8900.54] read 19739.86 [9062.25, 10429.04]
</span></p>
</blockquote>
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">But my actual Lustre performances are pretty poor in comparison (can't top 8GB/s write and 13.5GB/s read)<br>
So I started to question my lnet tuning but playing with peer_credits and max_rpc_per_pages didn't help.<br>
</span><br>
<span style="font-family:Helvetica,sans-serif">My test setup consist of 133x10G Ethernet clients (uplinks between end devices and OSS are 2x100G for every 20 nodes).<br>
The single OSS is fitted with a bonding of 2x100G Ethernet.<br>
<br>
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.<br>
<br>
Testing a single 10G client</span></p>
<blockquote style="margin-top:5pt;margin-bottom:5pt">
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">[LNet Rates of lfrom]</span><br>
<span style="font-family:Helvetica,sans-serif">[R] Avg: 2231 RPC/s Min: 2231 RPC/s Max: 2231 RPC/s</span><br>
<span style="font-family:Helvetica,sans-serif">[W] Avg: 1156 RPC/s Min: 1156 RPC/s Max: 1156 RPC/s</span><br>
<span style="font-family:Helvetica,sans-serif">[LNet Bandwidth of lfrom]</span><br>
<span style="font-family:Helvetica,sans-serif">[R] Avg: 1075.16 MiB/s Min: 1075.16 MiB/s Max: 1075.16 MiB/s
</span><br>
<span style="font-family:Helvetica,sans-serif">[W] Avg: 0.18 MiB/s Min: 0.18 MiB/s Max: 0.18 MiB/s
</span><br>
<span style="font-family:Helvetica,sans-serif">[LNet Rates of lto]</span><br>
<span style="font-family:Helvetica,sans-serif">[R] Avg: 1179 RPC/s Min: 1179 RPC/s Max: 1179 RPC/s</span><br>
<span style="font-family:Helvetica,sans-serif">[W] Avg: 2254 RPC/s Min: 2254 RPC/s Max: 2254 RPC/s</span><br>
<span style="font-family:Helvetica,sans-serif">[LNet Bandwidth of lto]</span><br>
<span style="font-family:Helvetica,sans-serif">[R] Avg: 0.19 MiB/s Min: 0.19 MiB/s Max: 0.19 MiB/s
</span><br>
<span style="font-family:Helvetica,sans-serif">[W] Avg: 1075.17 MiB/s Min: 1075.17 MiB/s Max: 1075.17 MiB/s
</span></p>
</blockquote>
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">With 10x10G clients :</span></p>
<blockquote style="margin-top:5pt;margin-bottom:5pt">
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">[LNet Rates of lfrom]<br>
[R] Avg: 1416 RPC/s Min: 1102 RPC/s Max: 1642 RPC/s<br>
[W] Avg: 708 RPC/s Min: 551 RPC/s Max: 821 RPC/s<br>
[LNet Bandwidth of lfrom]<br>
[R] Avg: 708.20 MiB/s Min: 550.77 MiB/s Max: 820.96 MiB/s <br>
[W] Avg: 0.11 MiB/s Min: 0.08 MiB/s Max: 0.13 MiB/s <br>
[LNet Rates of lto]<br>
[R] Avg: 7084 RPC/s Min: 7084 RPC/s Max: 7084 RPC/s<br>
[W] Avg: 14165 RPC/s Min: 14165 RPC/s Max: 14165 RPC/s<br>
[LNet Bandwidth of lto]<br>
[R] Avg: 1.08 MiB/s Min: 1.08 MiB/s Max: 1.08 MiB/s <br>
[W] Avg: 7081.86 MiB/s Min: 7081.86 MiB/s Max: 7081.86 MiB/s </span></p>
</blockquote>
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif"><br>
With all 133x10G clients:</span></p>
<blockquote style="margin-top:5pt;margin-bottom:5pt">
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">[LNet Rates of lfrom]<br>
[R] Avg: 510 RPC/s Min: 98 RPC/s Max: 23457 RPC/s<br>
[W] Avg: 510 RPC/s Min: 49 RPC/s Max: 45863 RPC/s<br>
[LNet Bandwidth of lfrom]<br>
[R] Avg: 169.87 MiB/s Min: 48.77 MiB/s Max: 341.26 MiB/s <br>
[W] Avg: 169.86 MiB/s Min: 0.01 MiB/s Max: 22757.92 MiB/s <br>
[LNet Rates of lto]<br>
[R] Avg: 23458 RPC/s Min: 23458 RPC/s Max: 23458 RPC/s<br>
[W] Avg: 45876 RPC/s Min: 45876 RPC/s Max: 45876 RPC/s<br>
[LNet Bandwidth of lto]<br>
[R] Avg: 341.12 MiB/s Min: 341.12 MiB/s Max: 341.12 MiB/s <br>
[W] Avg: 22758.42 MiB/s Min: 22758.42 MiB/s Max: 22758.42 MiB/s </span></p>
</blockquote>
<p class="MsoNormal" style="margin-bottom:12pt"><br>
<span style="font-family:Helvetica,sans-serif">So if I add clients the aggregate write bandwidth somewhat stacks, but the read bandwidth decrease ???<br>
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.<br>
<br>
This are my ksocklnd tuning :</span></p>
<blockquote style="margin-top:5pt;margin-bottom:5pt">
<p class="MsoNormal" style="margin-bottom:12pt"><span style="font-family:Helvetica,sans-serif"># for i in /sys/module/ksocklnd/parameters/*; do echo "$i : $(cat $i)"; done<br>
/sys/module/ksocklnd/parameters/credits : 1024<br>
/sys/module/ksocklnd/parameters/eager_ack : 0<br>
/sys/module/ksocklnd/parameters/enable_csum : 0<br>
/sys/module/ksocklnd/parameters/enable_irq_affinity : 0<br>
/sys/module/ksocklnd/parameters/inject_csum_error : 0<br>
/sys/module/ksocklnd/parameters/keepalive : 30<br>
/sys/module/ksocklnd/parameters/keepalive_count : 5<br>
/sys/module/ksocklnd/parameters/keepalive_idle : 30<br>
/sys/module/ksocklnd/parameters/keepalive_intvl : 5<br>
/sys/module/ksocklnd/parameters/max_reconnectms : 60000<br>
/sys/module/ksocklnd/parameters/min_bulk : 1024<br>
/sys/module/ksocklnd/parameters/min_reconnectms : 1000<br>
/sys/module/ksocklnd/parameters/nagle : 0<br>
/sys/module/ksocklnd/parameters/nconnds : 4<br>
/sys/module/ksocklnd/parameters/nconnds_max : 64<br>
/sys/module/ksocklnd/parameters/nonblk_zcack : 1<br>
/sys/module/ksocklnd/parameters/nscheds : 12<br>
/sys/module/ksocklnd/parameters/peer_buffer_credits : 0<br>
/sys/module/ksocklnd/parameters/peer_credits : 128<br>
/sys/module/ksocklnd/parameters/peer_timeout : 180<br>
/sys/module/ksocklnd/parameters/round_robin : 1<br>
/sys/module/ksocklnd/parameters/rx_buffer_size : 0<br>
/sys/module/ksocklnd/parameters/sock_timeout : 50<br>
/sys/module/ksocklnd/parameters/tx_buffer_size : 0<br>
/sys/module/ksocklnd/parameters/typed_conns : 1<br>
/sys/module/ksocklnd/parameters/zc_min_payload : 16384<br>
/sys/module/ksocklnd/parameters/zc_recv : 0<br>
/sys/module/ksocklnd/parameters/zc_recv_min_nfrags : 16</span></p>
</blockquote>
<p class="MsoNormal"><span style="font-family:Helvetica,sans-serif">Best regards,<br>
Louis</span></p>
</div>
<br>
<br>
<p style="font-family:Verdana;font-size:10pt;color:rgb(102,102,102)"><b>Disclaimer</b></p>
<p style="font-family:Verdana;font-size:8pt;color:rgb(102,102,102)">Please see our
<a href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.nag.co.uk_content_privacy-2Dnotice&d=DwMFaQ&c=KV_I7O14pmwRcmAVyJ1eg4Jwb8Y2JAxuL5YgMGHpjcQ&r=FTXmt89oLXmbXfP78w86-PxB1XdLYgxG8hEoAnZvCvs&m=ivu1XulCDlgfl0ZcF1MK057NBl_19awcsWYrT5l6Oc4&s=tESFOnq7ARkp3XH6U8CfNj1XnaIFjlOgyULJ0N8vyVs&e=" target="_blank" moz-do-not-send="true">
Privacy Notice</a> for information on how we process personal data.<br>
<br>
This e-mail has been scanned for all viruses and malware, and may have been automatically archived by Mimecast Ltd, an innovator in Software as a Service (SaaS) for business.</p>
</div>
_______________________________________________<br>
lustre-discuss mailing list<br>
<a href="mailto:lustre-discuss@lists.lustre.org" target="_blank" moz-do-not-send="true">lustre-discuss@lists.lustre.org</a><br>
<a href="https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.lustre.org_listinfo.cgi_lustre-2Ddiscuss-2Dlustre.org&d=DwMFaQ&c=KV_I7O14pmwRcmAVyJ1eg4Jwb8Y2JAxuL5YgMGHpjcQ&r=FTXmt89oLXmbXfP78w86-PxB1XdLYgxG8hEoAnZvCvs&m=ivu1XulCDlgfl0ZcF1MK057NBl_19awcsWYrT5l6Oc4&s=TUTMNQW1_S-T21CojVx-IpvMNY76NEsInuhtRTms770&e=" rel="noreferrer" target="_blank" moz-do-not-send="true">http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org</a><br>
</blockquote>
</div>
</blockquote>
<br>
</body>
</html>