[OpenAFS] Extremely poor write performance.

[Rubino Geiß]

> On 16 Jan 2003, Martin Schulz wrote:
> 
> > Paul Blackburn <mpb@est.ibm.com> writes:
> > 
> > > It seems that you looked at the performance of 1 client with 1 
> > > server? Are your figures for a single access to a server?
> > 
> > That is the natural measuring a user would do. That is what is 
> > important to *him*. That is what *he* will judge AFS on.
> 
> I don't give my users a server to themselves. I wouldn't if 
> it were NFS, either. 

True, but I'd like to provide adequate performance for my useres, so
accepting such a bottelneck as file creation is my setup, is no good at all.
Rigth? We should ask ouerselfs: whats the reason? 

I have tested the following: network saturation, server load. Non of them
are in critical numbers if you do mass deletes. So my clues are:	
	1. guess: maybe it's the network latency. 
		- contraidication: on 10Mbit we got the same numbers but 5
ro 10 times higher ping turnarounds. 
	2. guess: a flaw in the server code or linux.
		- guestion: how can I profile the fileserver? I want to know
where he is consuming time / waiting. How is this done for a client?

> > How does this affect the "delete 10000 small files" performance?
> 
> If you use a Solaris fileserver instead of Linux, file 
> deletion performance probably improves, for instance. There 
> are lots of variables.

We do not have access to Sparc hardware but intel-solaris, does this count,
too?

What else could help?

Bye, Ruby