[Info-vax] TCPIP Tuning

[Jeffrey H. Coffield]

One of our customers was subsumed by another larger company and their 
OpenVMS systems were moved to a new data center. In preparation for the 
move, the backups were ftp'd to a server in the Amazon cloud. At that 
point I had to change 3 tcpip parameters the get the transfer rate up. I 
set sb_max to 4000000 and tcp_sendspace and tcp_recvspace to 3000000 
(3M) each. This increased the transfer rate from about 3-4 Mb/sec to 
35-40 Mb/sec. My understanding that the effect of this would be 
increased non paged pool usage which I closely monitored and saw no 
problems.

The move was done and in the new setup, the production system would 
create savesets, ftp them to the Amazon server and then use ssh to tell 
their backup system to bring the savesets down and restore them so we 
had a hot spare. About once a week the backup server would lose access 
to the network and I had to stop/start TCPIP to get it to come back. I 
changed the backups so the restore didn't start until after the saves 
were all finished and the problem seemed to go away.

Two days ago the switch that the systems were connected to was replaced 
with a newer model and all looked good initially. The network people 
said the settings were the same and I verified that both the OpenVMS 
system and the switch showed 1000MB full duplex. I did some quick ftp 
checks and the transfer rate was good. The next morning the production 
system lost network access. Ping, telnet, ssh and web were all down. I 
connected through the management processor console, stopped the web 
servers in preparation to restarting TCPIP and the network came back. No 
need to stop/star TCPIP. This lasted for a few hours and then it hung 
again. Again I stopped the web servers and the network came back. 
Suspecting the tcpip changes, I reset them to the default settings and 
the system stayed up the rest of the day.

This morning the backup transfers were still running as the transfer 
rate was slow so I killed the transfers. Users still said the system was 
slow and after checking what I could I set the sb_max to 2000000 and 
tcp_sendspace and tcp_recvspace to 300000 (30k) and the system seemed to 
come back.

I do not have access to the switches but I know they are Cisco. Both 
OpenVMS systems are RX2800 i2 Itaniums running HP OpenVMS V8.4. Nonpaged 
Dynamic Memory is currently 142.66MB total with 106.72 free. The web 
servers only run on the production system and would only be started on 
the backup system in case of a manual fail over after changing ip addresses.

I have never had to change tcpip settings on any VMS system so this is 
new territory for me.

Any help/hints will be appreciated.

Jeff Coffield