nettune modification in SG cause TOC

[Anthony_4]

Hi, experts.
I am using 2 T600 in a cluster. I used nettune to make some modification as follow:
tcp_keepstart: from 7200 to 300
tcp_keepfreq: from 75 to 15
tcp_keepstop: from 600 to 300
But, after I run nettune successfully, the host suddenly rebooted. I check the OLDsyslog, it said "node timeout" and this node cannot join the cluster, SG TOC this node...
I don't think the heartbeat probe relates to these 3 tcp parameter.
any hints?

[melvyn burnard]

MC/ServiceGuard does not support the use of the nettune command especially when the cmcld daemons are already running.
If you really want to use the nettune/ndd facility, this MUST be done before the cluster is started.

[Anthony_4]

Thanks Melvyn
but could you pls explain this further?
i.e. if I run nettune before cluster is run, it should be ok?
is the 3 parameter I used ok?
Tony

[melvyn burnard]

The nettune commandsw are unsupported with MC/ServcieGuard at ANY time.

However, if you do wish to use them then ensure that you have them run BEFORE the rc3.d/cmcluster script is run.

This does not guarantee that you will not have a problem if the node is starting and about to join a running cluster.

[Stephen Doud]

Hi John,

If you have access to the ITRC knowledge database, this issue is discussed in document ID UXSGLVKBAN00000010 - "Why Did MC/ServiceGuard Initiate A TOC (crash)?"

As Melvyn said... definitely don't mess with the tcp timing after the node joins/starts a cluster.

Some customers have reported that using nettune before the cluster starts works.

-sed