MD raid speed going down....

Robert O'Kane (okane@em.uni-frankfurt.de)
Fri, 13 Mar 1998 05:20:24 +0100


Greetings,

I have been testing and useing the RAID5 md driver
with Linux-2.1.89-pre2 -> 2.1.90-pre2 and have found
my read rates as measured with bonnie going down by
HALF! The writes seem to be the same.

Actually, 2.1.89-pre5 was the fastest. (See below)
2.1.89-final is 1/2 the read speeds and 'random seeks'.
the 2.1Kernels are still 2x faster than the 2.0.3x
kernels.

The only thing which changes is the kernel.

Asus-P2T4 with IBM 6x86MX - PR233+
64Mb RAM
Libc-5.4.44
Linux-threads-0.71
(2x) Symbios-810A scsi cards with (4x)Fujitsu 4Gb drives
Useing the 'old' 53c7,8xx driver.
(Its MUCH faster than the 'new' ncr driver in the 89-pre5
kernel. 89-final and above both are the same speed).
Raid-tools-0.42-pre3

Anyone have an Idea? I could start hand patching stuff
but I would like to have a 'good' starting point!

Thanks,

Bob.

-------Sequential Output-------- ---Sequential Input--
--Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block---
--Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU
/sec %CPU
89-pre5 250 6030 70.1 11461 35.4 4933 25.7 6393 78.7 14585 29.3
164.5 3.9
89-pre5 250 5984 68.9 10100 28.4 4883 25.9 6438 79.0 14238 28.2
168.7 3.8
89-pre5 250 6006 70.1 11387 35.7 5095 29.2 6494 79.5 14647 29.2
154.0 4.5
89-pre5 250 5996 70.1 10971 31.7 4786 25.0 6345 78.0 14530 28.7
157.4 3.7
89-pre5 250 5930 69.2 11559 35.2 4951 27.4 6590 81.2 14619 29.9
176.9 4.6
----
89-final 250 5812 68.6 10620 30.7 3443 18.8 4459 55.3 6569 13.4
95.5 2.4
89-final 250 5956 69.8 10232 29.8 1762 9.3 5139 63.9 5178 10.9
88.4 2.2
89-final 250 5960 69.3 10760 32.0 1614 8.4 4821 59.7 6730 13.9
97.0 3.2
89-final 250 5996 69.9 10636 31.7 1821 9.3 5319 65.6 6188 12.9
87.0 3.8
89-final 250 5946 69.7 10834 31.0 2048 10.0 4712 58.7 5379 11.7
87.5 3.3
-----
90-pre1 250 5675 67.3 11230 35.2 3095 16.4 4539 56.4 6740 15.9
94.5 2.7
90-pre1 250 5959 69.3 10618 32.2 3063 16.7 4666 58.6 5722 13.2
90.4 2.8
90-pre1 250 6006 69.4 11233 33.1 1941 9.0 4445 55.2 5330 10.8
89.0 2.8
90-pre1 250 5982 69.5 10107 30.0 3057 15.6 4546 56.7 6647 14.4
86.9 2.3
90-pre1 250 5993 69.6 10902 34.2 2913 15.2 4966 61.8 6610 15.1
94.1 2.2
90-pre1 250 5968 69.2 10610 31.5 2989 15.8 5016 62.5 5608 13.3
89.4 2.5
-----
90-pre2 250 5864 69.4 10548 31.4 3023 16.1 5270 65.7 7012 15.1
87.0 2.2
90-pre2 250 5963 69.1 10882 32.6 1576 7.3 5294 65.5 6845 15.4
97.1 2.6
90-pre2 250 5992 69.4 10383 30.5 2524 12.5 4907 60.4 6845 14.8
91.2 3.1
90-pre2 250 5942 69.1 11431 34.0 2564 12.7 4021 50.5 7516 16.3
90.5 2.4
90-pre2 250 5977 68.9 10606 30.4 3069 16.3 5305 66.2 6245 13.2
87.0 3.2

-- 
*/
----------------------------------------------
Unfortunatly, I find language (like politics) necessary for humans.
Leider finde Ich Sprache (wie Politik) fuer die Menschen notwendig.
----------------------------------------------
WHO   : Robert O'Kane
WHERE : Künstlergruppe OTHERSPACE
WHAT  : Founding Member and Linker

WHY : http://www.rz.uni-frankfurt.de/~okane HOW : okane@em.uni-frankfurt.de ----------------------------------------------

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu