Skip to contentSkip to conversation

horrible disk thorughput on itanium

DISCUSSION Andi Kleen 5 participants 5 messages Dec 6, 2001 – Dec 7, 2001 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Andi Kleen

author
#1 lore raw
Quoted 4 lines · Andrew Morton

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.

The usual difference is if you have a pthreads capable C library or not. For newer glibc bonnie++ should definitely use putc_unlocked(); otherwise it'll eat lock overhead for each character to take the FILE lock.

As far as I can see bonnie++ doesn't use putc_unlocked, but putc.

With libc5 it likely would magically get a lot faster @)

-Andi

#2 lore raw
Quoted 6 lines · Andi Kleen
Quoted 4 lines · Andrew Morton

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.

Quoted 6 lines

The usual difference is if you have a pthreads capable C library or not. For newer glibc bonnie++ should definitely use putc_unlocked(); otherwise it'll eat lock overhead for each character to take the FILE lock.

As far as I can see bonnie++ doesn't use putc_unlocked, but putc.

Plain old Bonnie suffered from the same thing. I long ago made it use putc_unlocked() here because throughput was horrible otherwise.

-Mike

#3 lore raw
In reply to Andi Kleen

In article <[email protected]>,

Quoted 7 lines · Mike Galbraith
Quoted 4 lines

Andrew Morton <[email protected]> writes:

As far as I can see bonnie++ doesn't use putc_unlocked, but putc.

Plain old Bonnie suffered from the same thing. I long ago made it use putc_unlocked() here because throughput was horrible otherwise.

Oh, yeah, blame it on bonnie.

"Our C library 'putc' is horribly sucky"

"Well, then, use something else then".

Isn't somebody ashamed of glibc and willing to try to fix it? It might be as simple as just testing a static flag "have I used pthread_create" or even a function pointer that gets switched around at pthread_create..

"putc()" is a standard function. If it sucks, let's get it fixed. And instead of changing bonnie, how about pinging the _real_ people who write sucky code?

Linus

#4 lore raw

In article <[email protected]>,

Quoted 3 lines · Linus Torvalds

Isn't somebody ashamed of glibc and willing to try to fix it? It might be as simple as just testing a static flag "have I used pthread_create" or even a function pointer that gets switched around at pthread_create..

As the person who started this thread, I'll say that I'm willing to test new alternatives, Redhat engineers gave me a newer kernel to see if it helped (it didn't) and if someone can give me (or point me to) a glibc with better io I'm glad.

Right now I have to explain to my boss why my $4K pentium computers do io faster than my $20K itanium computer. And since our major software code is 3rd party, we can't rewrite the appilcation.

#5 lore raw
Quoted 6 lines · Greg Hennessy

In article <[email protected]>,

Quoted 3 lines · Linus Torvalds

Isn't somebody ashamed of glibc and willing to try to fix it? It might be as simple as just testing a static flag "have I used pthread_create" or even a function pointer that gets switched around at pthread_create..

Quoted 8 lines

As the person who started this thread, I'll say that I'm willing to test new alternatives, Redhat engineers gave me a newer kernel to see if it helped (it didn't) and if someone can give me (or point me to) a glibc with better io I'm glad.

Right now I have to explain to my boss why my $4K pentium computers do io faster than my $20K itanium computer. And since our major software code is 3rd party, we can't rewrite the appilcation.

As I understand it, they just told you that the benchmark you use gives uncorrect data on the itanium box. IIRC, you said that users were complaing about poor performance, so I think you should investigate on the application(s) you need instead of bonnie. Try and collect more real world data (if possible) or at least use some other benchmarks.

If it turns out that the (hi-end, I presume) application you're using it's both disk I/O bound and it actually uses putc(), then maybe the problem it's just using 3rd party applications without sources... all you have to explain to your boss is that you know what the problem is, and how to fix it, and that the "fix" is really easy (... if only you had the source). Add those 20K to the TCO of your closed source application.

.TM.

Signature · 5 lines
      ____/  ____/   /
     /      /       /			Marco Colombo
    ___/  ___  /   /		      Technical Manager
   /          /   /			 ESI s.r.l.
 _____/ _____/  _/		       [email protected]