Messages in this thread | | | From | "Stephen C. Tweedie" <> | Date | Tue, 15 Feb 2000 14:46:06 +0000 (GMT) | Subject | Re: elevator-starvation-4 (2.2.14 && 2.3.42) |
| |
Hi,
On Mon, 14 Feb 2000 18:00:22 -0800, Bruce Thompson <bruce@otherother.com> said:
> Then I'm confused. If request overrun has been addressed then how > can the oft quoted "cat /dev/zero > /foo/bar" trick cause a problem? > Am I completely missing how this messes everything up?
Because these two things --- IO scheduling and writeback flow control --- are done by completely different components of the system, and with completely different purposes.
Write-behind, left unchecked, can obviously fill the whole of available memory with dirty, pinned buffers. Flow-control on write-behind is primarily a VM problem: we have to prevent dirty data from overwhelming the caches. However, we may still quite legitimately find ourselves with many MB of dirty data pending in the write queues before write throttling kicks in.
That much dirty data can clearly overwhelm the fairness of a simple CSCAN elevator scheduler, even when write throttling is working correctly.
--Stephen
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |