Re: [PATCH] 2.5.21 IDE 91

From: rwhron@earthlink.net
Date: Sat Jun 15 2002 - 05:42:39 EST


>> > test the crap out of a kernel :).

>> Linux Test Project's runalltests.sh has occasionally triggered a bug.

> Is this still happening? What was the bug?

That's just a general suggestion with regard to kernel testing.
Recent 2.5.x hasn't had a problem completing LTP runalltests.sh.
I've used LTP to narrow down a couple early 2.4.x reiserfs bugs to
a specific test case. LTP has occasionally triggered oops and
livelocks too. It's a useful regression test.

>> 2.5 took a drop in dbench throughput recently.
>>
>> dbench ext2 128 processes Average High Low(MB/sec)

> Is this still with 384 megs of memory?

Yes.

> 2.5.19 18.60 21.69 14.58
> 2.5.20 12.89 13.15 12.79

> One possibile culprit here is the doubling of the request queue size
> in 2.5.20. A long time ago it was 1024 slots. Then it went to
> 128. That's where it is in Marcelo kernels. Then -ac kernels
> went up to 1024 because they have read-latency2. Somehow 2.5 found
> itself at 256 slots. In 2.5.20 it slealthily snuck up to 512
> slots. I didn't squeak about this because I was interested to see what
> effect it would have.

Interesting. I've seen read-latency2 drop dbench throughput in -aa
kernels (but I use it anyway). I'd like to capture the request
queue size. Is there a command or file in /proc that displays
it, or should I just grep drivers/block/ll_rw_blk.c?

> Does this patch get the throughput back?

I will try that next.

-- 
Randy Hron

- 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/



This archive was generated by hypermail 2b29 : Sat Jun 15 2002 - 22:00:33 EST