Slow NFS ?

All types of networks, network stacks, and protocols supported by OpenVMS.
Post Reply

Topic author
dante123
Member
Posts: 7
Joined: Thu Jan 07, 2021 10:55 am
Reputation: 0
Status: Offline

Slow NFS ?

Post by dante123 » Fri Apr 08, 2022 3:57 am

I'm experiencing poor performance with NFS
Is this "normal" or is there any parameter that can improve the performance ?

-----------
$tcpip sh ver
HP TCP/IP Services for OpenVMS Industry Standard 64 Version V5.7 - ECO 5
on an HP Integrity rx2800 i2 (1.60GHz/5.0MB) running OpenVMS V8.4-2L1
$TCPIP mount DNFS1: /host=serverxxx /PATH="NFSPath"

The server on this test is a windows server, but we got the same performance with a linux one


-----------
Here some measurement
when copying a big file from openVMS to the NFS share:
Network: EIC0(1000mbps) Last 10 sec [Avg RCV: 3196 pkt/s 2.161 mbps (0%) SND: 11200 pkt/s 125.676 mbps (13%)]

during backup with backup software:
Network: EIC0(1000mbps) Last 1 min [Avg RCV: 4747 pkt/s 3.622 mbps (0%) SND: 111542 pkt/s 850.999 mbps (85%)]

when a copy/ftp to another OpenVMS (not on same net-card but, there are no differences in network structure)
Network: LLA0(1000mbps) Last 10 sec [Avg RCV: 12311 pkt/s 6.014 mbps (1%) SND: 59734 pkt/s 689.944 mbps (69%)]


debbee.west
VSI Expert
Contributor
Posts: 12
Joined: Mon Oct 07, 2019 11:10 am
Reputation: 0
Status: Offline

Re: Slow NFS ?

Post by debbee.west » Wed May 11, 2022 3:44 pm

You should take a look at your NFS ACP parameters. You may want to increase the quotas for each ACP. Also, mount with /processor=unique.

If you are just doing backups, you may want to try this setting on the account running the backup:
Maximize RMS buffers for user account/process:
Place in LOGIN.COM for the user account doing the backup
$ set rms_default/block=127/buff=255/net=127/ext=5000

You should also take a look at your TCP settings. For NFS server, we advise to increase your TCP buffers:
Increase TCPIP buffers by editing TCPIP$ETC:SYSCONFIGTAB.DAT and add:

inet:
tcp_mssdflt=1460
tcbhashnum=16
tcbhashsize=16384
tcpnodelack=1
tcp_sendspace=1000000
tcp_recvspace=1000000
udp_sendspace=1000000
udp_recvspace=1000000

A system reboot is required for modifying tcbhashnum and tcbhashsize.

However, if a reboot is not immediately possible, the rest of the parameters can be set on the running system:

$ TCPIP
TCPIP> sysconfig -r inet tcp_mssdflt=1460
TCPIP> sysconfig -r inet pmtu_enabled=0
TCPIP> sysconfig -r inet tcpnodelack=1
TCPIP> sysconfig -r inet tcp_sendspace=1000000
TCPIP> sysconfig -r inet tcp_recvspace=1000000
TCPIP> sysconfig -r inet udp_sendspace=1000000
TCPIP> sysconfig -r inet udp_recvspace=1000000

Note the *_*space cannot exceed socket parameter sb_max which is 1048576 by default

Also, you may want to look at fastpath. BG0 and your ethernet device and your DNFS devices should be on the same CPU.

Depending on how many CPU's your system has, we may also need to look at settting Fastpath:
i.e.
$ SET DEVICE/PREFERRED_CPU=2 BG0
$ SET DEVICE/PREFERRED_CPU=3 EWA0
$ SET DEVICE/PREFERRED_CPU=4 DNFS0

And ensure no other devices/processes are preferring these CPU's:
$ SHOW CPU/FULL

if they are, move them to another CPU:
$ SET DEVICE/PREFERRED_CPU=<cpu_number> [DEVICE]

Now, the above tuning could result in more npagedyn usage.
$ ANALYZE/SYSTEM
SDA> CLUE MEMORY/STATISTIC
- monitor "Failed Alloc Requests "
SDA> CLUE MEMORY/FREE
- monitor "Largest free chunk:", if it approaches 0, increase NPAGEDYN and/or tune system.

User avatar

volkerhalle
Master
Posts: 196
Joined: Fri Aug 14, 2020 11:31 am
Reputation: 0
Status: Offline

Re: Slow NFS ?

Post by volkerhalle » Thu May 12, 2022 1:36 am

Dante123,

'performance' is a complex thing and many factors are affecting it.

Please clarify: you are talking about OpenVMS NFS client performance, not OpenVMS NFS server - right ?

The backup performance reported looks good - is it writing to the same NFS share as well ? Even the COPY/FTP performance looks acceptable (69% of interface speed). This at least shows, that the OpenVMS LAN driver can provide an acceptable network performance. And the TCPIP network drivers as well. Does the 'backup software' also work over TCP/IP, if so, then it seems to be well-tuned.

Did you try copying a 'big file' using the Linux NFS client to the Windows NFS server ? Performance ?

Did you increase the RMS parameters for the 'copy big file' case: SET RMS/BLOCK_COUNT=.../BUFFER_COUNT=.../EXTEND=...

Maybe try to first use TTCP for a performance test between the OpenVMS system (see SYS$SYSTEM:TCPIP$TTCP.EXE) and the Windows NFS server. This should show the achievable raw TCPIP throughput between those 2 systems without additional factors like hihger-level protocol used (NFS) and disk IOs.

Volker.
Last edited by volkerhalle on Thu May 12, 2022 3:16 am, edited 2 times in total.


Topic author
dante123
Member
Posts: 7
Joined: Thu Jan 07, 2021 10:55 am
Reputation: 0
Status: Offline

Re: Slow NFS ?

Post by dante123 » Thu May 12, 2022 5:17 am

I understand that it is complex ...

Yes it is the NFS client on OpenVMS that seems slow, (the server is not an OpenVMS server).
The backup and copy/ftp destnation is not on the same NFS. But uses the same network infrastructure.

I have not done any changes in configuration yet but I will try those given by debbee (incl. RSM)
and report here what was or not successfull.
Last edited by dante123 on Thu May 12, 2022 10:56 am, edited 1 time in total.

Post Reply