Q: I/O schedulers

[Ulrich Windl]

Hi!

I hope this mail will make it after kernel.org seems do be down for weeks now...

I have a question on I/O schedulers: I/O schedulers work for a single block device (more or less). Now I wonder: is there any attempt to accieve fairness across several block devices that use the same I/O channel?

The background is this: We have a huge SAN disk system that can deviver about 380MB/s through a 4Gb/s FC-Connection. One Linux system alone can pull that from the box. Now such a single Linux system uses some ten virtual disks from the SAN box. When a huge file (several GB) is copied from one SAN disk to another one the same box), at some point I/O gets very slow, and some commands seem unresponsive for several seconds (e.g. more than 60 seconds). Alltogether about 20 machines are connected to that SAN disk system.

Is there any attempt to give every block device using a specific HBA a fair share of the bandwidth? Also, what could be done in the current code to prevent such a situation. I suspect the problem are asynchronous buffer flushs combined with a lot of memory (> 100GB).

Please CC: any replies to me as I'm not subscribed to this list.

Regards,
Ulrich


--
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/