On 2007/03/29 22:55, Siju George wrote:
> On 3/29/07, Kyle George <[EMAIL PROTECTED]> wrote:
> >On Wed, 28 Mar 2007, Watson Crick wrote:
> >
> >> I've got OpenBSD 4.0 (release) on a laptop setup up as a router between
> >> 2 subnets, and providing internet access through a 3rd nic to a DSL
> >> modem. The problem is the bandwidth between the two subnets. I'm only
> >> getting a maximum of about 500 KB/s between two 100mbit cards. Top shows
> >> ~70% interrupt (~29% idle) while these transfers are going on. I don't
> >> know what the bottleneck is in the system. Are the Linksys PCMCIA nics
> >> crappy? Did I screw something else up?
> >
> >Try http://www.openbsd.org/faq/faq6.html#Tuning.
> >
> >Increase net.inet.tcp.{send,recv}space.
> >
>
> It says
>
> "You would normally use this to allow for routing or connection
> problems. Of course, for it to be most effective, both sides of the
> connection need to use similar values."
>
> If you have an ISP that gives you IP aadrees ( using PPPOE ) it there
> a way to measure or detect the valuse on the ISP's side?
The ISP don't normally have anything to do with this (excepting any
connections to their servers) (but see below about proxies). The relevant
settings are those on the endpoints of the TCP connection.
You might want to increase {send,recv}space if you have a connection
which has high bandwidth *and* high latency (i.e. ping times). But
it will only make a difference when you connect to servers which also
have high window sizes configured; often busy servers don't since it
increases the memory requirements.
If you're interested to see how altering this looks from the
perspective of network packets, run tcpdump(8) and watch how the
values in TCP SYN packets change as you vary the sysctl values
and make connections.
If there is a proxy in the path between you and the "real"
endpoint, the TCP endpoints are then your machine and that proxy.
In those cases, the ISP (or whoever) does have control over these
tuning parameters.
