NFS question

[MarkW_1]

We have an HP K570 running an Oracle database. It averages 95% mem usages(4 gig). There are 6/200 CPUs ave. 65%. We are wondering if Using NFS to mount the application and/or the databases drives with another box could help us relieve this system. Are there better alternatives?

[Jeff Schussele]

Hi Mark,

No, where the drives reside will have no impact on CPU usage. The binaries still have to run on the CPUs in the system. In fact, IF you setup the Oracle binaries OR data drives as NFS you will actually ADD cpu load to the system as it has to deal with all the PRC calls. Besides, I would definitely not recommend this at all as by default NFS uses UDP (non-error checking transmission protocol) for transport.

The only things that will lessen the load are:

1) More efficient Oracle routines & config.
2) More or faster CPUs
3) Kernel parameter tuning (if needed).
4) Faster System (L, N, etc.)

Rgds,
Jeff

[MarkW_1]

We are also thinking about upgrading to 64-bit os, 240 cpus, 8 gig ram. In doing so could we then use NFS?

[Jeff Schussele · Accepted solution]

No, I really don't think using NFS in this scenario is a good idea at all. First thing to consider would be redundancy. Even IF the system serving the NFS had mirrored drives for the binaries and/or data, there'd be NO redundancy on the network that delivers the data to you, the client. You lose that network - for whatever reason - you lose either the Oracle binaries or data.
The other major reason not to do this would be performance. Even if you ran this across a dedicated Gigabit ether network, you'll still not even be remotely close to the performance of a local SCSI disk - let alone a fibre channel disk. The network overhead will just kill performance.

If you think you're local disks are contributing to your performance problem - and you really should investigate why they may be - then you need to consider using a fibre channel array of some sort with multiple links such that you can spread the load across 2 or more controllers.

I just really think you should not even consider NFS in this scenario for any reason. I no of absolutely noone that uses NFS for real-time Oracle purposes.

Rgds,
Jeff

[Bill Hassell]

95% memory usage is good, not bad. You paid a lot for the RAM so why leave it empty? If your Oracle apps are running slowly, you likely need more RAM. But it all depends--you may have terrible SQL queries that are wasting time and disk I/O. Or the Oracle SGA is too small.

NFS is very unstable for production systems. The LAST thing you want is a busy network to completely lockup your server (and it WILL lockup). NFS offers no advantage for performance. To see what you need to know about NFS, buy the HP-UX NFS book.

There are no quick fixes when an app fails to perform as desired.

[harry d brown jr]

Mark,

How about fixing this issue:

This member has assigned points to 60 of 228 responses to his/her questions.

You can find your previous posts in this link to your profile:

http://forums.itrc.hp.com/cm/TopSolutions/1,,CA556749!1!questions,00.html

live free or die
harry

[Todd McDaniel_1]

First, let me tell you about this option for Oracle. It has its own buffer cache that you should be using instead of using the Host buffer cache in memory. This will speed up your writing to disk somewhat and free up your system memory that has been used for buffer cache for writing to disk for oracle DBs.

delaylog,nodatainlog,mincache=direct,convosync=direct 0 2

Also, let me say that memory use can be very high depending on reserved, allocated, lockable status. Your system should free up memory as needed by swapping out if it gets to 100%. Your scheduler should also kick in more often to page out or mark processes as scheduled to die or sleep to free up memory as needed.

This goes back to setting up and using the Oracle Buffer cache instead of the Host buffer cache.