Re: Nexus in DC?

Great advice Marc! And great tread by everybody. Read though, and I was
like, do I have anything meaningful to contribute... Well, maybe a word a
two. So here it goes..
Having been part of Cisco's EFT of the Nexus7k (we had it in our lab, and I
evaluated it for our upcoming DC projects) and having evaluated both the
Nexus 5K (with 2K FEX) and the Juniper EX4200, I have to say you you guys
hit all the key points. For us the fact the the Nexus7K 32port line cards
weren't line rate, we saw better value in other products (at the time, I'm
talking last year) what were better tested for our core such as the Catalyst
4900M or even back to Catalyst 6504 with (2) 6708 cards. Between cost, power
(yes the Nexus7K is a beast when it comes to power/cooling requirements),
and 10GIG density requirements (at the time, not over 16ports), we choose to
aggregate our 10GIG core with the Catalyst4900M. However, that was last year
early this year. Oh how time has changed, and so has our 10GIG density
requirements. So as we embark on new DC deployments, the Nexus7K (with
Cisco's promised upcoming line cards) are back in discussions as a possible
solution for our future new cores.

There are two things I shall mention regarding the Nexus5K.
Its a great Layer 2 (yes, no Routing on it yet) 10G aggregation core. But
without Layer3, you can't use it as a core replacement, for it must work
with some external Layer3 routing core. I think this is severly limiting the
deployment of the Nexus5K as a true independent 10GE aggregation point. I
believe Cisco is working on this (at least so they tell us), but I tell you
what, they've been numerous designs we've decided to go with the other
platforms (4900M, 6504 etc) just to eliminate a layer from our network. Why
this is the biggest issue in financial services.. "Latency Reduction".
Collapsed core and reducing the tiers is whats its all about nowadays to
squeeze every microsecond out of the network.

Finally my reviews on Juniper's strategy. Having only been through
presentations and never getting my hands dirty on the EX8200, I can't really
comment on this platform. But having lab tested (and now being deployed) the
EX4200, I can stay this is very good switch. But not in the same catagory as
the Nexus5K. Nexus5K is a high 10G Density, Layer2 switch. The EX4200 has
only 2 10GE ports and 48 1GE ports. To really compare a cisco platform to
it, would be best said that the Catalyst 4948 is its competitor. But the
EX4200 is a better switch that the 4948. Its an extremely low latent switch
(much betters numbers than the 4948), and it matches, and beats the 4948 in
feature set. It also can be stacked (I believe Marc mentioned this in his
"virtual chassis" statement which there way of saying its a stackable switch
like the Catalyst 3750). I could go on and on, but rest assured its an
impressive box. But not a competitor for the Nexus5K. I spoke to Juniper
reps, and they do have a product that should be coming out early 2010 that
is a direct competitor (Layer2, 10GE aggregation, FCOE) to the Nexus5K.

Anyway, great discussion guys! I learnt something. Hopefully I gave
information that someone can learn as well.

Adios.

- Abdul.

On Wed, Sep 2, 2009 at 3:11 AM, Marc La Porte <marc.a.laporte_at_gmail.com>wrote:

> Strategy is a big word I think. For my current project I have been in many
> meetings with Juniper and their equipment is simply not tailor-made for DC
> environments but are multi-purpose yet focused on SP environments (which
> means any features are always first produced for that market and not for
> DC)
>
> Think about this:
> # their EX4200 10-switch virtual chassis works best with a proprietary fat
> copper stack cable of 5-meters max (which doesn't scale well across
> multiple
> racks)
> # the alternative for the stack cable is using fiber, but this diminishes
> the backplane from 128-Gbps/FD to 20-Gbps/FD (uhm, the world upside down!!)
> # the propose the MX960 as DC core switch, but it's basically an
> over-dimensioned router for DC purpose
> # the 16-ports 10-GE blades for the MX are ridiculously expensive ($340K
> list!)
> # their DC solution (with VPLS) only works in a 2-tier design
> # they will come with a 10-GE access switch (like the Nexus5000) which can
> be put in any virtual chassis combination (pretty cool)
> # the "super-switch" is a matrix-like setup on the MX or EX8200 with shared
> backplane etc, but I fear for the type of interconnect
> # they do not have a strategy for 40GE ports
> # they do not have a strategy for integrating VMware
> # they do not have a strategy for FCoE / CNA
>
> Do I need to go on? ;-)...
>
> Just my 2ct.
>
> Cheers,
> Marc
>
> On Wed, Sep 2, 2009 at 08:09, Rick Mur <rmur_at_ipexpert.com> wrote:
>
> > They do have a strategy and are making a 'super-switch' which means
> > clustering all switches in your DC and even across DC's, it's still due
> to
> > be released and there is not much information available.
> >
> >
>
> http://www.juniper.net/us/en/company/press-center/press-releases/2009/pr_2009
> _02_24-12_00.html
> >
> >
> > --
> >
> > Regards,
> >
> > Rick Mur
> > CCIE2 #21946 (R&S / Service Provider)
> > Sr. Support Engineer IPexpert, Inc.
> > URL: http://www.IPexpert.com
> >
> >
> > On Wed, Sep 2, 2009 at 1:44 AM, Chris Riling <criling_at_gmail.com> wrote:
> >
> >> What is Juniper's storage / unified I/O strategy? Do they have an answer
> >> to L2 datacenter interconnect besides traditional spanning tree /
> EoMPLS?
> >>
> >> Chris
> >>
> >> On Tue, Sep 1, 2009 at 4:01 PM, Rick Mur <rmur_at_ipexpert.com> wrote:
> >>
> >>> Yeah those new DC projects are awesome I might say.
> >>> I'm also a fan of NX-OS. Much better CLI than IOS-XR :-)
> >>>
> >>> We are planning for a customer that requires 80!!! Gbps of internet
> >>> connection an entire network build on Nexus switches and UCS blades.
> Would
> >>> be so awesome!
> >>>
> >>> I will be going on a UCS bootcamp quite soon and implement the first
> two
> >>> chassis in Holland ;-)
> >>>
> >>> Cool stuff!
> >>>
> >>>
> >>> --
> >>> Regards,
> >>>
> >>> Rick Mur
> >>> CCIE2 #21946 (R&S / Service Provider)
> >>> Sr. Support Engineer IPexpert, Inc.
> >>> URL: http://www.IPexpert.com
> >>>
> >>> On 1 sep 2009, at 19:52, Iwan Hoogendoorn wrote:
> >>>
> >>> Hi Marc,
> >>>>
> >>>> I find this a very nice project.
> >>>> I have done a lot of reading on the Nexus technology and I really can
> >>>> not wait to put my hands on a Nexus Project.
> >>>> But then again its just with all types of expensive technologie ...
> >>>> your boss need to have he money of have customers who has the money
> >>>> and who needs this.
> >>>>
> >>>>
> >>>> --
> >>>> Regards,
> >>>>
> >>>> Iwan Hoogendoorn
> >>>> CCIE3 #13084 (R&S / Security / SP)
> >>>> Sr. Support Engineer IPexpert, Inc.
> >>>> URL: http://www.IPexpert.com
> >>>>
> >>>>
> >>>> On Wed, Aug 26, 2009 at 10:27 AM, Marc La Porte<
> >>>> marc.a.laporte_at_gmail.com> wrote:
> >>>>
> >>>>> Hey guys,
> >>>>>
> >>>>> Any of you already having practical experience with big Nexus
> roll-outs
> >>>>> in
> >>>>> Data Centers? Running into any "problems"?
> >>>>>
> >>>>> For a customer we are planning to deploy 5020 end-of-row switches
> with
> >>>>> 2148T
> >>>>> top-of-rack switches, aggregated in 7018s fully loaded with 32-port
> >>>>> 10-Gbps
> >>>>> blades. Just to give you an idea, the complete data center would be
> >>>>> around
> >>>>> 256 rows (30,000 ports)...
> >>>>>
> >>>>> Cheers,
> >>>>> Marc
> >>>>>
> >>>>>
> >>>>> Blogs and organic groups at http://www.ccie.net
> >>>>>
> >>>>>
> _______________________________________________________________________
> >>>>> Subscription information may be found at:
> >>>>> http://www.groupstudy.com/list/CCIELab.html
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>
> >>>>
> >>>> --
> >>>> Regards,
> >>>>
> >>>> Iwan Hoogendoorn
> >>>> CCIE #13084 (R&S / Security / SP)
> >>>> Sr. Support Engineer IPexpert, Inc.
> >>>> URL: http://www.IPexpert.com
> >>>>
> >>>>
> >>>> Blogs and organic groups at http://www.ccie.net
> >>>>
> >>>>
> _______________________________________________________________________
> >>>> Subscription information may be found at:
> >>>> http://www.groupstudy.com/list/CCIELab.html
> >>>>
> >>>
> >>>
> >>> Blogs and organic groups at http://www.ccie.net
> >>>
> >>> _______________________________________________________________________
> >>> Subscription information may be found at:
> >>> http://www.groupstudy.com/list/CCIELab.html
>
>
> Blogs and organic groups at http://www.ccie.net
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html

Blogs and organic groups at http://www.ccie.net
Received on Tue Sep 08 2009 - 08:14:56 ART