> Yes, I started looking through the lower-level code. I am puzzled though
> as to why the low-level sorting doesn't seem to have much effect in 2.0.33
> at least. Using a stock 2.0.33 kernel with write-caching turned off on the
> drive itself causes a huge performance hit. (I can post the benchmark program
> I used if anyone wants to try it themselves)
One new 2.1.x function in the sort code is attempt_merge() in
drivers/block/ll_rw_blk.c. I doubt that it accounts for the difference,
but perhaps -- we can try adding a printk() to that function, and verifying
whether it is called frequently under the benchmark.
Gadi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu