Skip to contentSkip to conversation

horrible disk thorughput on itanium

DISCUSSION Greg Hennessy 4 participants 4 messages Dec 6, 2001 – Dec 9, 2001

lore
Share & tools
lore
#0 lore raw

I recently installed a both a Dell dual cpu 2500 server (dual 1.6 ghz ia32 chips) and a dell 7150 (dual IA64 chips). My users complained that the disk io speed on the itanium seemed very slow, even though both servers have a megaraid controller with seagate cheetah disks. Bonnie also shows the ia64 machine having worse throughput than the ia32 machine.

[root@hydra bonnie]# cat bonnie.hydra bonnie.leo

              -------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
          100  1765 100.0 282891 100.1 377295 100.0  2058 100.0 592709
	  99.5 51920.4 196.5
              -------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
          100 17049 100.1 265197 101.0 197094 98.2 16631 100.4 675831
	  99.0 40400.0 191.9

Hydra is the itanium, leo is the 32 bit machine. The character io of hydra is a factor of 10 slower than that of leo. Is this more likely a kernel issue, or a glibc issue? Both machiness run standard redhat 7.1, and 2.4.9-12smp kernels.

#1 lore raw
Quoted 7 lines · Greg Hennessy

I recently installed a both a Dell dual cpu 2500 server (dual 1.6 ghz ia32 chips) and a dell 7150 (dual IA64 chips). My users complained that the disk io speed on the itanium seemed very slow, even though both servers have a megaraid controller with seagate cheetah disks. Bonnie also shows the ia64 machine having worse throughput than the ia32 machine.

How much RAM do you have ?

#2 lore raw
In reply to Greg Hennessy
Quoted 7 lines · Greg Hennessy

...
Hydra is the itanium, leo is the 32 bit machine. The character io of hydra is a factor of 10 slower than that of leo. Is this more likely a kernel issue, or a glibc issue? Both machiness run standard redhat 7.1, and 2.4.9-12smp kernels.

The character I/O part of bonnie++ writes a single character at a time, via stdio. It's more a test of your C library than of the kernel.

The fact that you get the same throughput on each platform with the block I/O part of the test indicates that the hardware and kernel are OK, but the C library is broken.

Not sure how to diagnose this. Probably you should write a simple five-line stdio-based test program, see if that exhibits the same behaviour, then fiddle with setvbuf().

-

#3 lore raw
In reply to Greg Hennessy

Hi,

Quoted 1 line · Greg Hennessy

[root@hydra bonnie]# cat bonnie.hydra bonnie.leo

[...]

Quoted 5 lines
	      --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU
/sec %CPU
          100  1765 100.0 282891 100.1 377295 100.0  2058 100.0 592709
	  99.5 51920.4 196.5

I bet you have more than 100MB RAM, so you measure memory performance instead of disk performance in the block reads and writes. Same for seeks. Don't do that. Always use at least twice you ram size if you're interested in seeing your disk speed.
Newer bonnies warn you about it. The current release is 1.2.
http://www.garloff.de/kurt/linux/bonnie/

The char reads/writes seem to suffer glibc overhead.
You can use the _unlocked variants with option -u and see whether this makes a difference.

Regards,

Signature · 3 lines
Kurt Garloff  <[email protected]>                          Eindhoven, NL
GPG key: See mail header, key servers         Linux kernel development
SuSE GmbH, Nuernberg, DE                                SCSI, Security