[c-nsp] bonded T1s into 7206VXR

Bill Wichers billw at waveform.net
Mon Feb 21 19:46:10 EST 2005


> OK, folks, here's another related question: we're trying to choose between
> these bonded T1's and a 5M fiber connection, for a collocation customer in
> our datacenter. The customer is slightly concerned that the bonded T1's
> will
> provide less effective bandwidth than the fiber connection, due to the
> 1.5M
> nature of each individual circuit.

Your customer is correct, the fiber 5 Mb/s will be better. With the bonded
T1s you'll get the same latency as a single T1 would have, just more
packets will fit down the pipe in any given second. For some applications
this matters. Also, the 5 Mb/s fiber can probably scale up to at least 10
Mb/s, probably 100 Mb/s (maybe more) easily, while the T1s won't -- at
least not without adding more T1s and their associated cabling.

> In my experience, T1's tend to get slow when I try to push >1.35M through
> them. I'm wondering if (3) bonded T1's will actually result in more like
> 4M
> of effective bandwidth.... From that p.o.v. a single 5M circuit over fiber
> might give slightly better performance.

IME, faster circuits can be loaded to higher percentages of total capacity
before people notice any performance degradation. I attribute this to the
lower latency, and the lower serialization delay allowing packets to
"squeeze through" with less delay since more packets fly by every second
on the faster pipes. Lower "interspertion delay", I suppose... We used to
have that problem between two POPs, and moving from MLPPP bonded T1s to a
single DS3 circuit completely solved all our problems -- including some
performance issues we'd never attributed to the MLPPP bundle.

BTW, multiple T1s don't really buy you much in the way of reliability.
Sure, you have multiple T1s and all, but chances are all of your T1s will
be inside the same multipair cable, use the same punch blocks, and
probably go into the same card cage in the carrier's CO. In fiber-fed
buildings I usually see one large fiber mux (usually Fujitsu FLM's in
Ameritech/SBC land) that feeds all the T1s in the building. If that FLM
goes down, or the fiber feeding it, then *all* the T1s go down. I wouldn't
count "redundancy" as a big factor in picking multiple T1's over the fiber
option -- the multiple T1s would protect against a single port failure in
your router, but would offer little to protect against the much more
common failure of in-building or outdoor cabling.

     -Bill


*****************************
Waveform Technology
UNIX Systems Administrator




More information about the cisco-nsp mailing list