[OpenAFS-devel] 1.4.8 performance

Hartmut Reuter reuter@rzg.mpg.de
Wed, 08 Oct 2008 17:47:15 +0200


After integration of my object storage stuff into 1.4.8pre1 I have made 
some tests between two of our cell's fileserver machines,
Both Quadcore Intel(R) Xeon(R) CPU           E5405  @ 2.00GHz
with 4 GB main memory. They are both on the same Gbit-Ethernet switch.

On the client I use 64 MB memory cache with chunk size 64 KB.

The filesystem on the server is rather fast for read (>300 MB/s), but
much slower for write (< 100 MB/s). Therefore performance for write is 
not really signifcant.

There are 5 different tests:

1) Normal write into AFS using the cache manager	~50 MB/s
2) Using "afsio" completely bypassing the cache manager ~70 MB/s
3) Normal read from AFS using the cache manager		~50 MB/s
4) Using "afsio" completely bypassing the cache manager ~83 MB/s
5) Normal read with Matt Benjamin's cache bypass       ~107 MB/s

This was a normal AFS-file (not OSD) of 8 GB size (so large to avoid 
buffer effects on server and client side).

The client machine was not used for anything else, but the server is
in normal production, however with low load.

I hope to get more disk space in the next days to set up also an OSD on 
the same switch. Then I will repeat these messages against rxosd.

-Hartmut
-----------------------------------------------------------------
Hartmut Reuter                  e-mail 		reuter@rzg.mpg.de
			   	phone 		 +49-89-3299-1328
			   	fax   		 +49-89-3299-1301
RZG (Rechenzentrum Garching)   	web    http://www.rzg.mpg.de/~hwr
Computing Center of the Max-Planck-Gesellschaft (MPG) and the
Institut fuer Plasmaphysik (IPP)
-----------------------------------------------------------------