Is useful to reset periodically the cluster (soft/hard) ?

[CAS_2]

Hi

[ First I want to remark that this is not a joke ]

For last one year, my company was implementing our project X in the headquarters of a client.
X is a system involving several software products (Oracle 9i, Service Guard, web servers) running on several different hardware (HA HP-UX servers, Windows Servers) and telecom gadgets (phone call mux, LAN switches, ...)

As X was not perfectly tunned or configured, there were a lot of manual soft/hard resets of servers, programs, LAN switches in order to solve some unexpected misfunctions. Reseting were always suggested by client to reduce the unavailable time of service.

After all this time, now the most of these problems were gone and no reset is needed. Our HP-UX servers run Oracle 9i for a lot of months with no long disruption of service due to Oracle nor HP servers.

But this client of ours is convinced that reseting the platform will make it to work better in order to prevent service disruptions. Resets would affect servers, laptops, disk arrays, LAN router, LAN switches...
and so on.

We're trying to take this idea out of his mind. We explained that these resets will provide no benefit and, in the other hand, they may cause damages.

My boss suggested to find a reliable document (by a reputed firm, University, ...) showing the resets will provide no benefit or the damages they may cause.

Do you know such docs ?

[Keith Bryson]

This is HP-UX - not Solaris!! I have 9i servers that have been up for 500 days without re-start. Solaris and its ever expanding kernel was the one that needed regular re-boots. Just keep your eye on defunt processes, process memory usage and process CPU usage. 'Top' can give you much of this info.

K

[Simon Hargrave]

Your system should not need bouncing just for the sake of it, as has been said it's possible to run for years without a reboot.

However you should really have a regular patching strategy for all your servers (whether you choose 6 monthly, annually etc is up to you), during which time you will need to reboot anyway.

The old adage of "if it ain't broke, don't fix it" is certainly true to an extent, however security patching plays a big part in driving the regular patching requirements. Also for example products such as Oracle always insist on a particular level of patching for support. If you come across a problem in your Oracle in 2 years time the first thing they'll check is your OS patch level, and unless you're reasonably up-to-date, they won't help you.

[Steven E. Protter]

The only reason I see for needing to bounce servers is to install patches or software that requires a boot.

Otherwise, you can run your boxes for a very long time without booting. Same goes for the cluster.

SEP

[Duncan Edmonstone]

I don't know of any documention that tells me not to stick my fingers in a power socket, but that doesn't mean I should do it! You won't find much in the way of documentation on this because the entire idea is just ludicrous.

I'd turn this on its head - ask your client what hard empirical evidence he has that resetting his entire environment will prevent service disruption. Ask him to point you at the documentation which states that the systems should be regularly reset. If all they can say is that 'they believe it will help' or 'they think it will help' then thats no reason at all - I could just as easily say 'I believe waving a dead chicken at the server will prevent service disruptions'!

This is a typical management reaction - it gives them something to do!

HTH

Duncan

[Geoff Wild]

Keith - 500 days with out a bounce! - man - you are way behind in your patching :)

I only reboot when patching, if NFS hangs, or if the applications are poorly written - and have memory leaks...

Now, for most of my Apps - they do come down during cold backups - something that I'm trying to convince my client it isn't needed - and online backups are fine.

No need to recylce the OS...as far as damages - that sounds like an old wives tale - rebooting a server doesn't cause damage...only causes an outage...

Rgds...Geoff