[131114] in North American Network Operators' Group
Re: Recommendations for Metro-Ethernet Equipment
daemon@ATHENA.MIT.EDU (Jason Lixfeld)
Wed Oct 20 15:57:26 2010
From: Jason Lixfeld <jason@lixfeld.ca>
In-Reply-To: <022801cb706a$ead9d5e0$c08d81a0$@net>
Date: Wed, 20 Oct 2010 15:57:19 -0400
To: nanog@nanog.org
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On 2010-10-20, at 11:24 AM, Eric Merkel wrote:
> Any suggestions, success or horror stories are appreciated. ;)
I've been going through pretty much the same exercise looking for a =
decent PE for almost two years. Our requirements were for a PE device =
that had between 12-24 ports (in a perfect world, mixed mode 10/100/1000 =
copper + SFP), 10G uplinks, EoMPLS, MPLS VPN, DHCP server, =
port-protect/UNI (or similar) capabilities, DC power and a small =
footprint (1RU)
Of all the ones we looked at (Juniper, Cisco, Extreme, Brocade, MRV, =
Alcatel) initially, MRV was the only contender. The rest either didn't =
have a product, or their offering didn't meet various points within our =
criteria.
As such, we bought a bunch of MRVs in early 2009 and after four months =
of trial and error, we yanked every single one out of the network. =46rom=
a physical perspective, the box was perfect. Port density was perfect, =
mixed-mode ports, promised a 10G uplink product soon, size was perfect, =
power was perfect, we thought we had it nailed. Unfortunately there are =
no words to describe how terrible the software was. The CLI took a =
little getting used to, which is pretty much par for the course when =
you're dealing with a new vendor, but the code itself was just =
absolutely broken, everywhere. Duplex issues, LDP constantly crashing =
taking the box with it, OSPF issues, the list went on and on. To their =
credit, they flew engineers up from the US and they were quite committed =
to making stuff work, but at the end of the day, they just couldn't make =
it go. We pulled the plug in May 2009 and I haven't heard a thing about =
their product since then, so maybe they've got it all together.
While meeting with Juniper a few months later about a different project, =
they said they had a product that might fit our needs. The EX4200. As =
such, we had a few of these loaned to our lab for a few months to put =
through their paces, from a features and interoperability perspective. =
They work[1] and they seem to work well. The show stopper was =
provisioning[1] and size. The box is massive, albeit it is still 1U.
[1] (I'm not a Juniper guy, so my recollection on specific terms and =
jargon may be a bit off kilter) they only support ccc, which makes =
provisioning an absolute nightmare. =46rom my experience with Cisco and =
MRV, you only have to configure the EoMPLS vc. On the EX4200, you have =
to create the LSPs as well. To get a ccc working, the JunOS code block =
was far larger and much more involved per vc than the single line Cisco =
equivalent. To create the LSPs was, I believe, two more equally large =
sized code blocks. At the end of the day, it was just too involved. We =
needed something simpler.
About the same time that we started to evaluate the EX4200, Cisco had =
pitched us on their (then alpha) Whales platform. It looked promising =
(MRV still had the best form factor) and we expressed our interest in =
getting a beta unit in as soon as we were able to. This is now known as =
the ME3600 and ME3800 platform and we've been testing a beta unit in our =
lab for the past few months. This is the platform we have chosen. It's =
not perfect, but our gripes have more to do with form factor (it's 1RU, =
but it's a bit deeper than what we'd like) and port densities (no mixed =
mode ports) than software or features. We've been pretty pleased with =
it's feature set and performance, but this hasn't seen any real world =
action, so who knows how that will turn out.
If you're asking more about a P router or P/PE hybrid, we've also just =
ordered a few ASR9000s under try-and-buy as P/PEs to close up the chains =
of ME3600s that will start to be deployed in our remote sites. A =
Juniper MX would certainly work well here too, and it seems to =
interoperate rather well with the ME3600s, so that's certainly an =
option, but for us, we think it will work more in our favor to go with =
the ASRs in the core, but if not, we'd ship them back under the =
try-and-buy and get Junipers instead.
Hope that helps.=
