[PVFS-users] PVFS API Performance with Multiple Processes

pvfstester at fastmail.com.au pvfstester at fastmail.com.au
Wed Dec 22 11:46:55 EST 2004


On Tue, 7 Dec 2004 08:59:56 -0600, "Nathan Poznick"
<poznick at conwaycorp.net> said:

> Thus spake pvfstester at fastmail.com.au:
>> I have three machines connected by a gigabit network, each acting as a
>> PVFS client and an IOD. Two machines have one 1GHz processor each and
>> 1GB of RAM; the third has two 3GHz Xeons with HyperThreading and 2GB of
>> RAM. Each machine has a single IDE drive which contains the operating
>> system, software, and the PVFS root directory. All machines are running
>> RedHat AS 3.

> It's a little unclear - are these 3 machines acting as both servers and
> clients?  If so, then performance can certainly be affected in new and
> interesting ways.  Historically, we see slightly higher performance when
> using the PVFS API than going through the VFS.

That is correct. Each of the three machines is both a client and a
server. (Though there is only one metadata server - the most powerful
machine.)

I have definitely seen better performance from the API than VFS on
simple tests (single-process program writing or reading large,
consecutive chunks). I was hoping to get the same gains by converting
this program to use the API, but I keep running into inexplicably slow
performance.
-- 
  
  pvfstester at fastmail.com.au



More information about the PVFS-users mailing list