On Mon, 23 Apr 2001, Jeffrey E. Harris wrote:
> On Sun, 22 Apr 2001, Fred Wright wrote:
>
> > Aside from allowing more machines, hubs have the benefit of providing LEDs
> > to show activity, which can be very handy when chasing problems. Activity
> > LEDs on stations, when they exist at all, are usually in some really
> > inconvenient place. If you *know* you'll be getting more machines later,
> > by all means use the hub from the start. 5-port hubs are only about $30
> > these days. The only disadvantage is the inability to use full duplex,
> > but it's pretty rare to get a significant benefit from that, anyway (plus,
> > older Amiga Ethernet cards typically don't support it).
>
> I bought a Linksys switch for about $40 after rebate, which supports
> 100Mb/Full Duplex. Unfortunately, the one test I ran at 100Mb/full
> duplex between two PCs (Windows based) resulted in a net transfer speed of
> about 1MB (yes MB) per second; stepping the post speed down to 10Mb/Half
> Duplex resulted in a transfer of about 500KB per second. I guess we get
> what we pay for!
Just because the hardware supports the speed doesn't mean the software can
run that fast. :-)
Does the switch have LEDs to indicate the speed and duplex mode of the
ports, and did those behave as expected?
On Tue, 24 Apr 2001, Carl Troein wrote:
> On 23-Apr-01, Jeffrey E. Harris wrote:
>
[...]
> > Duplex resulted in a transfer of about 500KB per second. I guess we get
> > what we pay for!
>
> Your cables were shielded, right? Have you tried with a single
> shielded (and crossed, of course) cable between the two PCs
> for comparison?
Shielding almost never matters with TP Ethernet unless the cable has to
run close to a large metal object or source of interference for a
significant distance.
> And how did you test the speed? On the Amiga (10Mbps) I've noticed
> a huge difference between different ftp clients and servers, just
> because some use asynch I/O and larger buffers. Oh, and on
> a 100Mbps link I suppose it might be useful to tweak the TCP
> window size and such - I've never tried that though.
Yes, you *might* need to crank up the socket buffer sizes to get full
100Mb speed with typical latencies, but you ought to be able to do better
than 1MB/s even without that. The rough rule of thumb is that you want
the socket buffer to be at least as large as the delay-bandwidth product
of the path. Since the RTT is probably no larger than a couple of
milliseconds in this case, a buffer size on the order of 25K should be
enough for 100Mb Ethernet.
Fred Wright
--
To unsubscribe send "unsubscribe miami-talk-ml" to
"[EMAIL PROTECTED]". For help on list commands send "help" to
"[EMAIL PROTECTED]".
