WORKSHOPS

Making ATM Infrastructure Decisions

Normally, when Network Computing decides to review products, it does so by defining a narrow product niche and then inviting all of the players within that niche to participate in testing. In the nearly six years we've been doing things this way, we've found this to be the most reasonable process for evaluating products. Readers receive a detailed and fair comparison, and vendors usually don't complain too bitterly about unfair treatment.

However, when the University of Wisconsin's Computer-Aided Engineering Center (CAE) set out to pick a company to provide equipment for a backbone u pgrade, we didn't l ook at the problem in the same way as Network Computing editors would when doing a review. My last official act before leaving the University of Wisconsin recently was to scout the ATM market and recommend vendors the CAE would use during the next three or four years.

Setting The Scene It is important to note that Cisco Systems has been our choice for high-end networking gear for some time now. Since we've been happy, it was up to the other vendors to show why they were significantly better. We say significantly because we already had a lot invested in Cisco in terms of our staff knowledge and understanding of the equipment. If the alternatives weren't demonstrably better, we would take the path of least resistance and stick with C isco.

We'll try to keep our Cisco bias out of this article, which in fact won't be too difficult. The four vendors we gave strong consideration to were Cisco, 3Com Corp., Bay Networks and Cabletron Systems/FORE Systems. Note that Cabletron OEMs its ATM line almost ex clusively from FORE; Cabletron made the pitch to us, however.

What We Wanted Our network is currently served by a FDDI ring with servers attached directly to the ring and workstations attached to Ethernet ports on our routers (see "CAE's FDDI Ring," this page). Our routers are five Cisco AGS+es and our Ethernet distribution is handled mostly by Cabletron equipment. Our plan was to move from FDDI to ATM, replacing routers, server NICs and some end-distribution equipment with newer ATM gear (see "CAE's Proposed ATM Network" on page 170). We fully intended to build a LANE 1.0 network as well, so all equipment quoted to us was LANE 1.0 compliant.

Specifically, we were looking for two or three ATM switches with a total of 30 OC-3 ports. We also needed six or more edge devices with a total of 60 to 100 Ethernet ports. We were curious about the expandability of the switches and edge devices and were particularly interested in their ability to support fast desktop technologie s such as Fast Ethernet or ATM. We also needed at least one router that would route between virtual LANs (VLANs) and provide a path back to the campus FDDI ring. The price quotes we received for the routers were staggering. Putting a FDDI module and an ATM module in a high-performance router is expensive.

In general, we asked for quotes that included equipment, management software and maintenance for at least 12 months. Even though we follow this market pretty closely, we were surprised at the variability of the offerings, both in terms of price and functionality. A summary of our findings can be seen in "CAE Infrastructure Project" on page 168.

Cisco At the time we started our search, the Cisco Lightstream 1010 was a big secret and the Catalyst 5000 was just getting FD DI and ATM ports. Cisco is still working on fully integrating its acquisitions of Kalpana and Grand Junction Networks. Putting ATM modules into the Kalpana boxes in particular looked like a winner for us since we have some smaller bu ildings where using the Cataly st 5000 would be cost prohibitive.

In its final specification, Cisco recommended the Lightstream 1010 as the ATM switch and the Catalyst 5000 and Catalyst 3000 as edge devices.

We liked several things about the Lightstream 1010. The daughterboard technology, which Cisco says it plans to use to support new ATM standards, makes a lot of sense to us. There definitely will be new standards that could render expensive switches useless. Cisco's approach seems a good hedge. The 1010 can hold as many as 32 ports and can support a number of ATM technologies, including OC-3 and OC-12--the technologies we were most interested in.

The Catalyst 5000 is a great box because it can support all of the media that we might want to use including Ethernet, Fast Ethernet, FDDI and ATM. Our only complaint with the 5000 is that it was a little pricey. Nonetheless, it has a lot of room for expansion and appeared--in our testing, at least--to be a great performer.

The Catalyst 3000 is one of the Kalpana switches with an AT M interface. It has 16 fixed ports and, on a price-per-port basis, is much less expensive than the 5000. We haven't tested it for performance, but if it does at all well, it should work fine for our less demanding sites.

In terms of the final price, one thing to note is that high-end Cisco routers are expensive. For ATM switches, the Lightstream 1010 is a bargain. The Catalyst 5000 is the only slotted packet switch (as opposed to cell-based switch) in the roundup, and that pushes the total price up for Cisco's edge switches. The 5000 is a very flexible box and we would consider it to be worth the extra cost.

Bay Networks There is no doubt that Bay was served well by its acquisition of Centillion Networks. Although we talked about the 5000 AH, Bay's final proposal involved only the Centillion 100. As it turns out, the Centillion 100 is quite a flexible box, supporting Ethernet, Token-Ring and ATM. There is a definite appeal to using the same box throughout the network. Maintaining spare parts i s easier and configuration obviously will be consistent throughout the network.

Depending on your attitude toward end stations using ATM, the Bay solution provides the ability to supply ATM to all points on the network. The Cisco design would have restricted us to using ATM at the backbone. At this time, we aren't looking to run ATM to the desktop. Private Ethernet and Fast Ethernet seem to be more economical ways at this point to get performance to end stations.

Because we could put ATM throughout the network, Bay's solution lends itself to creating a mesh network that should be fairly immune to fault. Of course, with all of our servers connected to one switch, there will be a single point of failure. And, if your fiber is laid like ours, you are subject to the errant-backhoe syn drome. Nonetheless, we could build a netwo rk that in many ways is more fault-tolerant than our FDDI network is today.

Bay has indicated that it will bring new form factors to the Centillion line, making the technology easier to implement off the backbone. The company will do this by making a smaller version of the Centillion 100. Although using the Centillion 100 everywhere is appealing for the sake of redundancy and the ability to deliver ATM throughout the network, its use also makes the network upgrade expensive. While providing the fewest total Ethernet ports, Bay's proposal was the most expensive in terms of total cost for edge and ATM switches. On the other hand, Bay provided the most ATM ports in its proposal.

Bay's routers are not cheap either. Bay and Cisco clearly understand who is on top of the route r market and have set their prices accordingly.

3Com With one of the most complete lines, 3Com has a lot of products at its disposal to fill a request like ours. The CellPlex 700 0 can take Ethernet modules just as the Bay Net works Centillion product could. 3Com has fixed-configuration switches that compare well to the Catalyst 3000. And, although 3Com doesn't have a box quite like the Catalyst 5000, the LANPlex product family offers some promise.

The obvious difference between 3Com and the others was the lower total cost. 3Com's routers are much less expensive than either Bay or Cisco. CellPlex 7000 is competitive with the Bay and Cisco offerings, and the fixed-configuration LinkSwitch 2700 makes the proposal very economical. You don't get the flexibility that you'd have by using the Catalyst 5000 or Centillion 100 as edge devices, but 3Com's overall proposal is tempting--it gets the job done and saves significant money.

Using the 3Com router is a bit of an unknown, at least as far as we're concerned. From what we've seen, it should perform adequately, but we'd want to test one before we bought it. That's true for all these proposals; price and flexibility are just part of the puzzle.

Cabletron By far the most unusual-- and proprietary--proposal was Cabletron's. For the most part, Cabletron OEMs equipment from FORE Systems. If you aren't a Cabletron shop to begin with, it makes more sense to deal directly with FORE. This is even true now that FORE has purchased Alantec and therefore competes for some of Cabletron's business.

The 200BX is a direct OEM of FORE's ATM switch. It's a good switch with a competitive price. The ESX-1320 is a fixed-configuration, Ethernet-to-ATM edge switch. It doesn't have as many ports as the Catalyst 3000 or the LinkSwitch 2700, yet it remains competitive on a cost-per-port basis.

What really makes Cabletron's proposal different is its suggestion to use Cabletron's Route Server technology. This is essentially the SecureFast technology that Cabletron has be en talking about for some time. If going with 3Com's proposal seemed risky to us, this was akin to betting the farm. It might have worked just fine, but it would have taken a lot of in-house testing to convince us.

Cabletron' s technology is a bit like Multi-Protocol Over ATM (MPOA) in that routes between subnets are determined right at the edge switch. That has the potential to become a great way to do things, but at this point it is highly proprietary.

Cabletron's equipment can also be used in the LANE 1.0 environment that we originally proposed. In this case, a Cisco router typically would be used to route between VLANs built within the ATM and edge switch fabric.

If you trust Cabletron's SecureFast switching, then its proposal looks to be cost-effective. If you want a LANE 1.0 environment, it may make more sense to deal directly with FORE.

The Lessons There are a few important lessons to be taken from this exercise. The first is that flexibility comes at a price. When Cisco and Bay propose flexible, slotted chassis boxes rather than the fixed-configuration edge switches from 3Com and Cabletron, the price increases considerably.

The second lesso n is that providing routing between VLANs and connectivity to our legacy FDDI network is a significant expense. ATM and FDDI interfaces for routers are very expensive. We wanted two routers for redundancy and that made the routers one of the most expensive parts of the project.

The third lesson seems to be that Cisco, 3Com and FORE (although we didn't request a proposal from them) have the most complete product lines for organizations looking to build an ATM infrastructure. We were eager to deal with just one vendor and these vendors have the best options to choose from. Bay certainly is catching up quickly--it's just a matter of integrating the Centillion technology into more of its line before it muscles to the top.

Cabletron's divergence from the rest of the pack, with its SecureFast technology and its reliance on FORE for ATM technology, is a b ig gamble. Although Cabletron is developing some of its own ATM technology, you'd better be pretty happy with the company as your sole vendor before consi dering it as your back-end technology provider.

Art Wittmann is the director of technology implementation at Evolve Software.

He can be reached at awittmann@nwc.com.