[151654] in North American Network Operators' Group
Re: Force10 E Series at the edge?
daemon@ATHENA.MIT.EDU (Brant Ian Stevens)
Wed Mar 28 11:47:38 2012
Date: Wed, 28 Mar 2012 11:46:32 -0400
From: Brant Ian Stevens <branto@networking-architecture.com>
To: George Bonser <gbonser@seven.com>
In-Reply-To: <4F733142.1010109@argentiumsolutions.com>
Cc: NANOG <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
> Brant Ian Stevens <mailto:branto@argentiumsolutions.com>
> March 28, 2012 11:41 AM
> The CER is the perfect box for this application, save for the
> redundant processors. The MLXe will work great if you want a small
> form factor and redundant processors.
>
> -Brant
> George Bonser <mailto:gbonser@seven.com>
> March 28, 2012 11:34 AM
>
>
> I have been using a pair of CER (but not the -RT) at one location for
> a while now and so far have been flawless. These particular units
> aren't taking full tables so don't need the -RT but I wouldn't have
> any trouble using them. The -RT are basically a 1U XMR.
>
> Tom Daly <mailto:tom@dyn.com>
> March 27, 2012 11:59 PM
> Brent,
> Your options include, for smaller boxes:
>
> - Brocade CER series, but make sure you the -RT versions due to RAM
> (haven't tried, though)
> - Juniper MX (MX80 is working well for us)
> - Cisco ASR1006 (heard a lot about BGP price issues)
>
> But for 300mb/sec, what not OpenBSD + Quagga?
>
> Tom
>
>
>
> ----- Original Message -----
>
> Jo Rhett <mailto:jrhett@netconsonance.com>
> March 27, 2012 6:00 PM
> I was very happy with the E300 as a data center core switch handling
> multiple full feeds (around 15) with about 10x the traffic you are
> talking about. The only problem I had was that Force10 didn't have a
> useful (basically forklift) upgrade to get more IPv4 prefixes, and the
> more I talked to them and the more I showed them the graphs
> demonstrating what we'd need for prefix space assuming even the most
> conservative assumptions at depletion, the more I realized they really
> Did Not Get It. In fact, their brand new architecture recently
> announced had only 500k prefixes allowed, at a time that the Juniper
> MX platform handled 2million easily.
>
> So I would be fine using Force10 again, given the following changes:
> 1. Large limits on IP prefixes allowed
> 2. Reallocation of useless memory from stupid things like MAC tables
> to prefixes (data centers have very few MACs, very many prefixes)
> 3. Command line logging
>
> The units worked great at failover, never had any problems gracefully
> failing over from one RP to another, but if you have to cold boot them
> for any reason it takes like 5 minutes :(
>
