[TAG] MTR over a CDMA link (cell phone)

[Jay R. Ashworth]

On Mon, Apr 05, 2004 at 10:56:54PM -0400, Ben Okopnik wrote:
> On Mon, Apr 05, 2004 at 10:01:36PM -0400, Jay R. Ashworth wrote:
> > On Mon, Apr 05, 2004 at 07:27:53PM -0400, Ben Okopnik wrote:
> > > > I have Verizon CDMA via Earthlink, and I usually don't drop any packets
> > > > in their cluster... but the best ping times I've *ever* seen are about
> > > > 400ms; 800 is more common.  CDMA ain't that fast.
> > >  
> > > Now that I don't have to take off running, I've figured it out: if I set
> > > the "re-ping" times to greater than the worst-case delay, I get 0% loss.
> > > 1 second wasn't enough.
> > 
> > It actually keeps track of which packets are outstanding; it will trend
> > the loss % to 0 if it gets everything, even if it's late.
>  
> Didn't do it for me, over the course of 100 pings.

I did say trend.  :-)  The loss % goes back down, when late pings show
up.

> > > > But thanks to windowing, that latency (heh heh) doesn't adversely
> > > > affect the available bandwidth as much as might seem obvious.
> > > 
> > > Run that by me again, please? I'm familiar with the concept of windowing
> > > in, say, ATM, but I'm not sure how it applies here - particularly since
> > > I know of no way to resolve a latency problem (you can combine
> > > low-bandwidth links to get more bandwidth, but adding more high-latency
> > > links will not improve latency.)
> > 
> > No, but windowing will keep latency from affecting bandwidth.  The
> > thing that impacts bandwidth, on a TCP stream, is the combined
> > latencies of packet transmission, packet turnaround at the receiving
> > end, ack transmission, and ack processing at the local end.
> 
> [Nod] Right, my mistake in interpreting "latency" this time (duh!
> You'dathunk, by now...)

And now you understand why *I* had trouble before.

> > The more packets you're permitted to have in the window, the more
> > bandwidth you can get out of a given amount of total latency... at the
> > price of a bigger stack of dominos hitting the floor if you don't have
> > a perfect link.
> 
> Sure; now I gotcha. Packet latency affects small chunks of data much
> more than it does large ones, of course.

Well, by percentage, yes, but it's the *empty* hose that kills you on
long-latency pipes.  Windowing allows you to fill up those holes.

> > Damn.  I'd thought you were.  I'll look at it someday, but it's damned
> > inconvenient not to be able to mail the output to someone... and since
> > it clears the screen on exit, you can't even copy and paste
> > conveniently.
> 
> So, why can't you just copy and paste it while it's running? I just
> tried it, and there doesn't seem to be any evil magic about that would
> prevent it.

I don't have good enough aim to do a complete select in the
less-than-one-second I have available between updates... which seem to
interrupt Konsole's select process.

Cheers,
-- jra
-- 
Jay R. Ashworth                                                jra at baylink.com
Member of the Technical Staff     Baylink                             RFC 2100
The Suncoast Freenet         The Things I Think
Tampa Bay, Florida        http://baylink.pitas.com             +1 727 647 1274

        "They had engineers in my day, too."  -- Perry Vance Nelson