[AusNOG] Juniper vs Cisco vs Brocade - what's best for BGP routing?

[Julien Goodwin]

On 11/12/13 22:43, Chris Ricks wrote:
> We accept a few domestic-only transit tables and the memory usage for
> these tables is surprising given my past encounters with other routers -
> I've seen Cisco 7206 boxes handle 2 full transit tables and a few IX
> tables in the same memory as the Juniper gear handles 2 domestic-only +
> default route tables.

There's a whole bunch of reasons for this (Cisco's BGP is much more
efficient for a start, it had to be back when RAM densities were much
lower), but for a start JunOS (by default) stores several copies of the
various tables. In practice this doesn't really matter for most people.

> Juniper gear (in our experience) tends to do one thing very well. By
> that I mean that if a JunOS instance is tasked with one of switching,
> security or routing the outcomes are good. Stacking two of these
> functions can be problematic and seemingly requires regular reboots of
> devices (JSRP is really a non-negotiable for firewall and/or routers -
> VC is the same for switching).  That said, the JSRP failover for BGP
> isn't perfect, and you should expect to see convergence time on failover.

Er, the fact you say "JSRP" means you really aren't using their routers,
only the SRX. Your routers should fail over with BGP & OSPF/ISIS first
and if you must use NSR between their own control cards only after that.
A single chassis, or for that matter cluster, is still a single point of
failure and to be avoided as far as possible.

Certainly the whole point of the MX line (when it launched at least) was
to mix switching and routing better than other platforms out there.

While the SRXs are fine, and a 650 could handle the transit needs of
many on this list for those of us that use the M/MX/T series boxes that
Juniper built their name on they're not "real Juniper's", and I'd much
rather use an MX5 for BGP despite its control plane issues.

Also other than a few issues that pop up at over a year uptime on an old
EX release I've not had any uptime-related issues with JunOS in years.

> One concerning outcome we have seen is the current fracturing of the
> JunOS feature base. The "single OS" promise Juniper promotes is
> currently problematic - we have multiple JunOS devices fulfilling
> switching, routing and firewall duties and the disparity of recommended
> JunOS versions between them has grown over the last 12 - 24 months.

This isn't too bad if you can at least keep to one release per series,
but I agree they've been heading down a disappointing path here, and
seem blinded to it in my conversations with them.