HP Requesting SWAP = RAM
Michael Weber
mweber at alliednational.com
Tue Apr 27 12:47:46 UTC 2010
Good morning!
If memory serves, the kernel goes into emergency memory saving mode when it detects the available swap space is less than 50% of physical ram. It does this to keep from crashing should memory usage go up. There is a pretty steep performance hit when it does, as you've found out. The kernel really does want swap => ram. It likes its elbow room. Even if you have enough memory that you should never need to swap out a memory page, you still have to have the swap space available.
Swapping the drives is one way to get the space needed, or you could stay with the iSCSI space. This assumes that you really will never need to use the swap, because iSCSI speed isn't near what DAS is. If there is even a remote chance your memory usage will go up, grab the drives now.
IMHO, of course!
Michael Weber
Network Administrator
Allied National, Inc.
4551 W . 107th St.
Suite 100
Overland Park, KS 66207
913-945-4313 is my direct number
>>> "Hearn, Stan J." <stan.hearn at nscorp.com> 4/27/2010 6:36 AM >>>
We have a new HP ProLiant BL460c G6 with 64GB RAM and mirrored 146GB drives running HP Network Node Manager 8.13.
We are having issues with the embedded PostreSQL software.
First HP said their software was not supported on RHEL 5.4, only RHEL 5.2. But I refused to downgrade and only offered to downgrade a defined list of packages. But that list was never provided.
Then they asked that we modify vm.overcommit_memory=2, but that didn't fix the problem.
So HP told us that 18GB of swap was not large enough, we needed at least 64GB of swap to equal the physical RAM.
Our usage statistics show less than 32 GB RAM ever being used and 0% of swap being used.
However to humor them we added an iSCSI target of 46GB to swap and the problem was resolved.
HP support writes "... it is not a matter of the space showing up as used. When our software poll command runs, It looks for a certain amount of swap and our experiences show that the recommended OS setting usually works and fixes the issue."
HP also referred to an old statement found on the linux.com "The general rule of thumb for swap size is that your total available swap space should be around double your RAM size. ..." I guess we should be thankful they are not asking for 128 GB of swap. At this same site I found many other sane answers to how much swap for a 64GB system.
The system owner is now wanting to replace the 146GB drives with 300GB ones so that we can create this huge swap file. This effects several systems.
Why does this seem wrong to me? I don't feel that I have much power to push back. I can't take out memory to reduce the swap space needed. I don't want to continue to run iSCSI on swap. I'm afraid if I relent all my users are going to want really large swap files for illegitimate reasons.
Help, give me some ammunition or options? Anyone else having vendors require ridiculous amounts of swap? Anyone else running HP NNM with swap issues?
Thanks,
Stan
--
redhat-sysadmin-list mailing list
redhat-sysadmin-list at redhat.com
https://www.redhat.com/mailman/listinfo/redhat-sysadmin-list
E-MAIL CONFIDENTIALITY NOTICE: This communication and any associated
file(s) may contain privileged, confidential or proprietary information
or be protected from disclosure under law ("Confidential Information").
Any use or disclosure of this Confidential Information, or taking any
action in reliance thereon, by any individual/entity other than the
intended recipient(s) is strictly prohibited. This Confidential
Information is intended solely for the use of the
individual(s) addressed. If you are not an intended recipient, you have
received this Confidential Information in error and have an obligation
to promptly inform the sender and permanently destroy, in its entirety,
this Confidential Information (and all copies thereof). E-mail is
handled in the strictest of confidence by Allied National, however,
unless sent encrypted, it is not a secure communication method and may
have been intercepted, edited or altered during transmission and
therefore is not guaranteed.
More information about the redhat-sysadmin-list
mailing list