Following the recent discussion about AMI`s rather dreadful
SMP performance I made the upgrade
from the 0.96 driver (included in 2.2.6) to 1.0.
This system has 6 * 9GB Atlas II drives, three per channel, setup as
write-back, 64K stripe
size, RAID 5:
Driver 0.96
-------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
1000 2915 80.0 8030 23.9 4148 22.7 3467 88.3 9856 23.6
93.1 4.7
Driver 1.0
-------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
1000 2792 70.3 9118 27.1 4836 26.1 3490 84.6 11499 27.3
244.0 12.5
Using the same drives in RAID 0 + 1 format (with 8K strip) yields
slightly worse
performance all round. Now, given that I want RAID for reliability, I
have no complaints
about the AMI card (no data lost so far), I bought AMI cards for the
price (so I`m still
happy on that count), but the performance is, umm, lame.
If you don't recognise the above benchmark layout, the program is
Bonnie. Using a
large Sun and software striping with dismal little 4GB drives, has
yielded bonnie values
several times the above - so I'm not too happy with this performance.
Anybody doing any better ?
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
