We are in the process of updating our internet connection to 8 t1's bound
together. Due to price, our options have been narrowed to AT&T and MCI.
I have two questions:
1. Which technology is better for binding t1's: multi link frame relay
(mci's) or load balancing (att's)
2. Which company has a better pop in Atlanta: mci or att?
We are in the Chattanooga TN area and our current connection is 6 t1's
through att but they will only bond 4 so they are split 4 and 2.
Bryce Enevoldson
Information Processing
Southern Adventist University
We are in the process of updating our internet connection to 8 t1's bound
together. Due to price, our options have been narrowed to AT&T and MCI.
I have two questions:
1. Which technology is better for binding t1's: multi link frame relay
(mci's) or load balancing (att's)
of course, as always... not mci's view on the world 
depends on what you want... do you want more than a 1.5mbps flow to pass?
or do you just want to get 9mb of bandwidth and you don't care about max
flow size? The MFR stuff will allow your link to look like a 9mb path, not
6 1.5mb paths. The load balancing makes it look like 6 l.5mb paths.
2. Which company has a better pop in Atlanta: mci or att?
i'll avoid this question since I'm not equiped to answer as anything but a
marketting answer 
We are in the Chattanooga TN area and our current connection is 6 t1's
through att but they will only bond 4 so they are split 4 and 2.
Some folks have said in the last that over 6mb of bandwidth it might be
better/cheaper/easier to just get a fractional/burstable DS3 to meet your
needs.
-Chris
It depends on what you mean by better. Multilink is more CPU-intensive
but is nicer to such things as voice that don't deal well with packets
arriving out of order. Load balancing can be per packet or per destination
(or flow). Per-packet allows for aggregation of the multiple paths for a
large flow between two specific points but can give voice and similar
services problems with reassembly.
So "better" will depend on the nature of your traffic.
At that speed I would highly recommend a DS-3 instead of either of the above.
One other consideration is the quality and reliability of the local loops. Does your local telco manage to keep T-1's running well? DS-3's? If you have multiple circuits, you may wind up being able to ride through equipment failures in the local loops (or not, depends a lot on how the loops are built).
I'm not arguing one or the otehr is better for reliability, but it's something to ask before signing a contract.
FWIW, I asked about MLFR on the cisco-nsp and one person responded
indicating he worked
for some time with Cisco on trying to get MLFR to work but never resolved
out of order and dropped packet
issues. What hardware are you considering using to do MLFR ? Please let me
know how things go as MLFR
would be a nice option on some parts of my network.
James H. Edwards
Routing and Security Administrator
At the Santa Fe Office: Internet at Cyber Mesa
jamesh@cybermesa.com noc@cybermesa.com
http://www.cybermesa.com/ContactCM
(505) 795-7101
In my experience the breakeven point for a Frame Relay DS3 is 6 DS1
circuits. DS3's tend to be more reliable than DS1's as the ILEC usually
installs a MUX at your site instead of running to the nearest channel bank
and running the T1's over copper with a few repeaters thrown in for
good measure.
Another nice thing about DS3's is that it is easy to scale bandwidth in
the future by modifying the CIR on your link. Another feature is that
since the link is faster the serialization delay is lower which will give
you better latency and last but not least PA3+ for Cisco 7[2|5]xx routers
are inexpensive and give you one call for service not a separate call for
the CSU/DSU's and the serial line card you need to support a multilink
solution.
Scott C. McGrath
Scott McGrath wrote:
In my experience the breakeven point for a Frame Relay DS3 is 6 DS1
circuits. DS3's tend to be more reliable than DS1's as the ILEC usually
installs a MUX at your site instead of running to the nearest channel bank
and running the T1's over copper with a few repeaters thrown in for
good measure.
I'll second that. Our ILEC extended our existing SONET node (for the PBX in another building) to our machine room (couldn't push DS3 over copper that far). Now, if they'd just terminate the old T1s at the new node and not push them over local copper from there to the machine room, we would be sitting pretty.
Another nice thing about DS3's is that it is easy to scale bandwidth in
the future by modifying the CIR on your link. Another feature is that
since the link is faster the serialization delay is lower which will give
you better latency and last but not least PA3+ for Cisco 7[2|5]xx routers
are inexpensive and give you one call for service not a separate call for
the CSU/DSU's and the serial line card you need to support a multilink
solution.
Ditto. We have one in a 7204 with a CIR of 30Mb. Handles it quite nicely, replaced 5 T1s on load-sharing per-packet link.
Jeff
3 quick notes--
Neither MLFR/FRF.16 (MCI's implementation) nor the corresponding CPE require
external DSUs. The service may utilize internal DSUs (whether on Cisco CPE
or Tasman) just as a tiered/fractional DS3 would.
ATM-IMA could be considered wasteful of bandwidth as you would have to live
with the ATM cell tax reducing usable bandwidth by about 25%. FRF.16 allows
for much lower overhead through frame relay encapsulation. FRF.16 also
allows for losing circuits within a bundle or even designating a threshold
number of circuits for when to consider a link "down" (useful in failover
scenarios).
Another minor point is that DS3s are tiered by many large providers through
timing at the provider edge DSU/linecard vs. CIR (even though FR encaps may
be used).
Given all that, a fraction DS3 may still be a better option if the telco
loop is reasonable...
Bryant Rump
Advanced Internetworking
Booz Allen Hamilton
rump_bryant@bah.com
Date: Thu, 16 Sep 2004 16:44:48 -0600
From: james edwards
FWIW, I asked about MLFR on the cisco-nsp and one person
responded indicating he worked for some time with Cisco on
trying to get MLFR to work but never resolved out of order
and dropped packet issues. What hardware are you considering
I must have missed your post, or I'd have chimed in. I've noted
MFR flakiness (dropped packets, sudden interface down, additional
links become unbonded and refuse to bond again, MTU troubles with
MPLS, wedged processes, etc.) on 12.3 and 12.3T.
Bryce, you might also consider an M13 mux if worried about loop
reliability. However, at this point, I'd place my bets on a DS3
loop as opposed to several MFR-ified T1s. :-/
Eddy