>>>>> " " == Riley Williams <rhw@memalpha.cx> writes:
>> - Actual network is congested, and you are running out of
>> socket buffer space. You'll probably need to wait for linux-2.4
>> before a general fix is available (it has already been included
>> in linux-2.3.x).
> The network in question consists of four PC's on a 10baseT
> network, one being a dedicated print server and another a
> dedicated modem server and firewall, neither of which generate
> much in the way of traffic. Those two are at opposite ends of
> the net segment, with the total length being about 10 metres,
> so there shouldn't be any length related problems either.
I wouldn't be so sure: this case depends largely on the network cards'
ability to quickly send off any request you put in the buffer. My
experience is that even a 100Mbit card can quickly build up a backlog
in the case of bulk transfers (which presumably is why people have
developed Gigabit cards). This turned out to be a major problem when
developing the write-gathering server optimized NFS client even on a
(by modern standards) slow PPro 180MHz.
Of course the robustness of the card's driver is also important, and
this often contributes to the problem (cf. the eepro100 driver under
linux-2.3.x where x < 40ish).
Cheers,
Trond
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Mar 07 2000 - 21:00:14 EST