 | | | No swap being used | No swap being used 2006-01-12 - By Jeff Macfarland
Back
I figured somone would bring this up. The raided partition is run over
fc-al to an enclosure that has no raid controllers, so hardware raid is
not an easy option at this point. I suppose that is an avenue I might
have to address even though I have my doubts about it being the problem.
Thanks for the help though. If I find it to be the problem, I will
report back here sometime.
William Warren wrote:
> aha..software raid. I bet that's the issue. You're simply reaching the
> limits of the system. Can you throw a good hardware raid controller in
> there and get your RAID off the cpu(which is where software raid resides).
>
> Jeff Macfarland wrote:
>> William Warren wrote:
>>> It depends on kernel tuning but most times the kernel will start
>>> dumping the file cache to free up ram. Once that is done it will
>>> start swapping. You pretty much have to chew up all ram before swap
>>> stats getting used. If you are having high iowaits then your issue
>>> is hdd or IO not swapping. Check your RAID controller or you may
>>> just need to upgrade to faster disks.
>>
>> We're pretty sure its IO as well but are keeping our eyes and ears
>> open. We have a case open with Sun about the slowdowns (sunrays are
>> severely affected, but ssh prompts seize up briefly as well)
>>
>> The 14 disks in question are 15k rpm 36G FC drives in an old compaq
>> enclosure. Its an older loop device with mdadm handling raid.
>>
>> I remember back in the day when 2.4.2x kernels were new where I had to
>> use some of Ingo Molnar's (sp?) latency patches or else desktop usage
>> was pitiful if the machine was doing any sort of IO. I've asked about
>> this IO problem before on this list (and elsewhere) and go no replies
>> so I figured it was just us experiencing this issue and not for lack
>> of kernel latency patches.
>>
>> IIRC, however, iostat shows high await times on /dev/sda whereas the
>> cvs repo is on the fc enclosure (/export) by way of a symlink on /.
>>
>>>
>>> Jeff Macfarland wrote:
>>>> Jussi Silvennoinen wrote:
>>>>>> Sun V20z, RHEL 3u5, i386 kernel, 12GB RAM, 12GB swap
>>>>>>
>>>>>> This box hosts quite a few things, among them serving SunRay's. We've
>>>>>> been experiencing horrible slowdowns particularly during a cvs
>>>>>> checkout.
>>>>>> Just noticed that NO swap is being used:
>>>>>>
>>>>>> output of `free`:
>>>>>>
>>>>>> total used free shared buffers
>>>>>> cached
>>>>>> Mem: 12277844 12036764 241080 0 1333552
>>>>>> 7502204
>>>>>> -/+ buffers/cache: 3201008 9076836
>>>>>> Swap: 12578704 0 12578704
>>>>>
>>>>> I guess yer a Solaris-admin? Things work a bit different in the
>>>>> penguin-world.
>>>>>
>>>>> No need to swap = nothing gets swapped. Solaris does it a bit
>>>>> different.
>>>>> Over 7gb's of your ram is used for block cache so you're far away
>>>>> from actually needing to swap.
>>>>>
>>>>
>>>> At what point will linux start using swap? Where there is absolutely
>>>> no free RAM left? Its been my understanding that swap will almost
>>>> always be used for idle processes regardless of amount of free ram
>>>> left. I see the output of free range from a few gigs free to 50M and
>>>> at no point is anything swapped. I suppose I am just used to our old
>>>> solaris boxes though.
>>>>
>>>>> Instead, look at the output of 'top' when the slowdowns occur.
>>>>
>>>> We have. IO waits perecentage goes quite high and iostat shows
>>>> 'await' time on the order of a few thousand ms for sda. The cvs repo
>>>> is on a software raid partition though.
>>>>
>>>> Thanks for the swap info though.
>>>>
>>>
>>
>>
>
--
Jeff Macfarland (jmacfarland@(protected))
Nexa Technologies - 972.747.8879
Unix Administrator
GPG Key ID: 0x5F1CA61B
GPG Key Server: hkp://wwwkeys.pgp.net
--
Taroon-list mailing list
Taroon-list@(protected)
https://www.redhat.com/mailman/listinfo/taroon-list
|
|
|