topic Re: Minimum failover time for SGLX in Operating System - Linux

Minimum failover time for SGLX

klemerS — Tue, 21 Mar 2006 07:56:44 GMT

Hi.

How can i determine The minimum failover time i can get - when a member fail or when a pkg fail? ( including the time its take to find hardware failare. )

its very simple application - start in notime&
killed in notime. ( almost)

I need 12 sec. for failover. Is it A
reasonable request??

Thanks.

Re: Minimum failover time for SGLX

Steven E. Protter — Tue, 21 Mar 2006 08:24:16 GMT

Shalom,

It is reasonable for most software, though oracle may give you trouble, since it often takes more than 12 seconds for an oracle instance to get started up.

You need to test this to be certain.

SEP

Re: Minimum failover time for SGLX

melvyn burnard — Tue, 21 Mar 2006 12:10:21 GMT

One thing to be aware of is that the failover time of a Serviceguard package can be determined and controlled by the various parameters, but the point to stress here is that it only makes the time taken to START the package deterministic. What is difficult to give a time for is the actual startup time of the application in the package, as this can be influenced by far too many factors, not least of which is roll forward or roll back of transactions for example.
So although you could configure for a time of 12 seconds from failure to the package start BEGINNING, you cannot realistically configure for a time whereby the application is up and ready (i.e. the package start has completed) for any and all scenarios

Re: Minimum failover time for SGLX

John Bigg — Wed, 22 Mar 2006 04:15:28 GMT

I do not believe that you will be able to get a 12 second failover time. Without the fast failover product which is available on HP-UX but not for Linux, with a 2 second node timeout you would expect around a 30 second failover time. If you reduce the node timeout to 1 second and have a 0.5 second heartbeat interval you can roughly halve this but I would think that puts your cluster at too much risk of transient problems resulting in false failovers.

And of course these times are simply the Serviceguard failover times and do not take into account any application startup or recovery which would be needed after Serviceguard failsover and starts the applicaitons on an adoptive node.

Re: Minimum failover time for SGLX

klemerS — Thu, 23 Mar 2006 02:59:02 GMT

Thanks for your replays.

I also think that 28 sec is the min i can get with QS & 56 sec with lock lun. I will

check it today.