Re: Multi-threaded IO performance regression on 2.6 kernel?

[Miquel van Smoorenburg]

In article <1784BBD8D1F15B4C9FB0F09F0A939F9001A3E463@xxxxxxxxxxxxxxxx>,
Xie, Bill <bill.xie@xxxxxxx> wrote:
>All,
>
>I am testing the multi-threaded IO performance on Opteron servers. 
>
>I use dd as the test tools. The single dd can reach 60MBps for single disk.
>
>on 2.6.5 kernel, If dd numbers exceed the CPU numbers, vmstat bi reduced
>to 20MBps.
>
>on 2.4.21 kernel, multi-threaded IO performance works fine, even I run
>40 dd command at same time.
>
>Does anybody experienced similar issue also?

First, use a more recent kernel, 2.6.11.x or 2.6.12.x. Then try if
changing the I/O scheduler (/sys/block/sda/queue/scheduler) to
deadline or cfq makes a difference.

I'm using 2.6.11.x and the cfq scheduler on most servers right now.

Mike.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/