On Fri, 5 May 2000, Thomas Charron wrote:
> Any large downloads or uploads to the machine gradually degregate
> bandwidthwise, untill they hit a timeout.. It's not the bandwidth to
> the MACHINE that degrades, just accross that oner individual socket.
On Sat, 6 May 2000, Thomas Charron wrote:
> Application really doesn't matter, any app transfering large
> amounts of data has this issue..
>
> After more research, I believe that the problem does not rely on the Linux
> box itself (I plugged my machine into their hub, and observed the same
> issues), but in the sDSL router/hub box that they are using. It provides
> NAT forwarding to the Linux box itself. Anyone have experience with these
> beasts, and what may be happening?
Could be a lot of things. (Sounds like your car mechanic, doesn't it? :)
First, try you tests under with many different machines using as many
different OSes and applications as possible, to rule out any possibility the
problem is coming from a particular host.
Could be the NAT box has a bandwidth choke on it, to prevent any one
connection from swamping the feed.
Could be the upstream ISP. I've heard stories about ISPs who allow more
bandwidth initially (so interactive surfing is fast) but choke longer
connections to keep people from leaching more then they are paying for.
Cable Internet providers in particular are notorious for this.
There could be a bug in the NAT box that is leaking resources or simply not
allocating them properly over time. Make sure it has the latest firmware,
etc., from the OEM. If possible, dump the black box NAT solution and go with
something Open Source.
Try rebooting (power cycle) the NAT box as a short-term fix.
Try using a network analyzer like tcpdump or ethereal to see if you can
find any clues on the wire.
--
Ben Scott <[EMAIL PROTECTED]>
| "Meddle not in the affairs of sysadmins, for they are quick to anger, and |
| have not need of subtlety." |
**********************************************************
To unsubscribe from this list, send mail to
[EMAIL PROTECTED] with the following text in the
*body* (*not* the subject line) of the letter:
unsubscribe gnhlug
**********************************************************
