[PlanetCCRMA] kjournald configuration

[Rick B]

Juan Reyes wrote:

>Rick, 
>Thanks a lot for your advice and tips. Looks like you are right. I tried
>changing the values of virtual memory with the kernel-tuning tool on the
>gnome-system tools menu and the situation seems to improve a bit.
>
>Of course I decided to switch from a 2.4 kernel to a 2.6 to see if the
>situation would improve. Funny - but things are the same. Looks that
>these default values are for servers and not workstations. I am going to
>look further for documentation about vm managing and if I could not find
>enough I might bother you for some pointers. 
> 
>  Cheers,
>
>  --* Juan
>
>On Tue, 2004-11-09 at 11:22, Rick B wrote:
>
>  
>
>>Juan,
>>    Kjournald probably isn't your problem and if it is you wouldn't 
>>necessarily want to extend the time it writes journals to disk you would 
>>probably need to shorten it. If you wait 30 seconds to write the journal 
>>to disk that just means that there is going to be that much more to 
>>write to disk, thus creating longer disk write times. But I don't 
>>believe that's your problem as I've struggled with this issue before and 
>>tuning my virtual memory has always solved the problem. By tuning vm to 
>>write to disk more often in smaller chunks it will help if not cure your 
>>problem. How to tune vm is dependent on which kernel you are using, the 
>>2.4 is easier to me because there is more documentation. With the 2.6 
>>kernel it's harder as the powers that be switched how vm works on the 
>>2.6 kernel and the documentation is minimal. By the way if your using a 
>>2.6.9 kernel, there was a bug in kswapd that caused excessive disk 
>>thrashing, but I'm sure Fernando would have patched that by now :-)
>>
>>                    Rick B   
>>    
>>
>
>
>  
>
    There are more tunable values in /proc/sys/vm. However finding the 
optimal value for each is a science akin to voodoo, as even the kernel 
maintainers get into flame wars over the correct value for 
"/proc/sys/vm/swappiness". Here's a link to one such debate:

http://kerneltrap.org/node/view/3000

                Rick B