High load average mail server 5.3-RELEASE

[Mariano Benedettini]

I've found on discussion lists that some people also tested values near 
80 or 100. I think I have CPU and RAM to start with a value of 50.
The rsize and wsize values are both 32768.

Thanks in advance,
Mariano.

Eric Anderson wrote:
> Mariano Benedettini wrote:
> 
>> Thanks for all the replies. It's not a HD problem.
>> On monday I'll increase the number of nfsd processes and the number of 
>> nfsiod on the client, setting both to 50,
>> I think that the nfs performance will be much better :-)
> 
> 
> 50 nfsiod's may be a bit overkill, but you should experiment to find out.
> 
> You should also increase the rsize and wsize parameters on the mount 
> options for better efficiency.
> 
> Eric
> 
> 
> 
>> Eric Anderson wrote:
>>
>>> Francisco Reyes wrote:
>>>
>>>> On Tue, 13 Sep 2005, mariano benedettini wrote:
>>>>
>>>>> 91.3% idle
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> CPU is not the problem. :-)
>>>>
>>>>
>>>>> Mem: 1599M Active, 1704M Inact, 311M Wired, 189M Cache, 112M Buf, 
>>>>> 14M Free
>>>>> Swap: 2023M Total, 184K Used, 2023M Free
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> Swap is not the problem.
>>>>
>>>>
>>>> Do
>>>> vmstat 10
>>>>
>>>> Watch the output.
>>>> In particular look at the first 3 columns.
>>>>  procs
>>>>  r b w
>>>>  1 1 0
>>>>  0 1 0
>>>>  1 1 0
>>>>
>>>> The left most column is CPU, the second column is disk IO.
>>>>
>>>> If you have a number in the "b" column and it never hits 0 you have 
>>>> an I/O problem. You HDs are not catching up.
>>>>
>>>> If you are using NFS and the "b" colun is not high and hits 0 
>>>> some/all the time then the bottleneck is either the nfs connection 
>>>> or the nfs server.
>>>>
>>>> For example I have some servers that the "b" column would be between 
>>>> 20 and 60 for a while. I am currently working on removing some of 
>>>> the load of the machine. In my case more memory would help, but the 
>>>> computer vendor we bought the machine from has sent us the wrong 
>>>> memory 3 TIMES!!
>>>
>>>
>>>
>>>
>>> Also, if it is an NFS server, one should check the cpu times on the 
>>> nfsd processes.  I've found that many times there aren't enough nfsd 
>>> processes to take the load from many clients.  Increasing the number 
>>> (double it) often helps this.  The max in 5.3 is 20, but you can 
>>> easily change it and get around it.
>>>
>>> Eric
>>>
>>>
>>>
>>>
> 
>