Re: 3Com 905C very slow (400Kbps)

On Wed, 4 Jun 2003, Florian Hackenberger wrote:

> Wow...I archive 2.1 MB/sec (ca. 45MB / 21sec)...ok, I believe your 
> theory about the disk beeing slow, but still...that is not 100Mbs.

Because the card, even if it does itself checksumming of the packets, does 
not host a complete IP stack - processing IP packets takes some time and 
this is where the slow CPU/mainboard/memory come into play. As I said 
prreviously, I have no idea about what figure should one expect from such 
hardware. But I would say to still try to get a transfer rate for purely 
network load (= without disk access also on the P4).

> Ok, so will go and buy a new IDE-Controller and will about get 2MB/s.

Keep in mind that this will also stay on the PCI bus and thus might again 
not reach its "normal" transfer rates,: it might recognize your disk as 
f.e. ATA100, but the real transfer rate might be just a small fraction of 
the theoretical maximum, just like for the networking. So it might be that 
adding that IDE-Controller will decrease the disk-to-disk transfer rate 
compared with the one described above which is only disk-to-memory.

> Since I've copyied the file once before, it seems to be in the cache of
> the computer reading it from the disk.

Yes, at this small size it's most likely that the disk was not used in the 
second (and subsequent) read.

> I don't know if the NFS client caches files too, but if not, it would
> mean that the hub cannot be the problem.

If you continue to do measurements using NFS, at least do it properly: 
unmount the FS on the client after each transfer. This will make sure that 
the client does not have anything cached.

-- 
Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: [email protected]

_______________________________________________
vortex mailing list
[email protected]
http://www.scyld.com/mailman/listinfo/vortex