Messages in this thread | | | Subject | Re: [PATCH] 2.5: push BKL out of llseek | From | Robert Love <> | Date | 29 Jan 2002 21:37:57 -0500 |
| |
On Tue, 2002-01-29 at 21:21, Dave Jones wrote:
> did you benchmark with anything other than dbench ?
No, and I really don't want to hear how dbench is a terrible benchmark. I didn't craft the patch around dbench and I think, here at least, dbench is an OK benchmark. I ran it numerous times over multiple client loads.
I think its clear there won't be a negative impact, because:
- acquiring the inode semaphore isn't any heavier (in the acquire case) than the BKL
- the lock contention on each inode semaphore is relatively zero
- besides just scaling badly with the using a global lock against all inodes, we use the BKL which in such workloads is already highly contested.
That said, I did do some lock profiling and latency tests. Contention was near-zero, but I only did 2-way testing. Under the preemptible kernel, while running dbench, scheduling latency improved 8.9%.
Robert Love
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |