HSRP availability in datacenters?

Fri May 11 20:41:16 UTC 2007

While I'm not a huge fan of running more than 32 instances on a 3550, using
the FAQ posted earlier to get above 16 works quite well.  

I'm not following the argument about failing 16 vlans at a time because
they're in the same group.  Running a quick test in the lab, this wasn't my
experience at all.  I'm not aware of the group instance having any
synchronization impact (such as it would with VRRP) when it comes to HSRP --
only a single vlan interface failed over when I did a shut on the primary.
The group simply determines the virtual mac address, but if I'm wrong on
this let me know.

The documentation/configuration synchronization issues are really more an
issue of how refined provisioning is.  If your upstream links from these
aggregation devices are layer 3, and I hope they are, the vlans carry only
locally significance anyway.  When the aggrs are spun up, the vlan
interfaces and groups could all be pre-defined before they're even needed.
Yes, you may not know the IP addresses or block sizes to pre-configure all
of the HSRP data, but you can hold the "standby x authentication" line
within a configuration without knowing any of the layer 3 information.  At a
later point when the vlan interface is actually needed for a customer, the
provisioning group simply needs to match the group number they already see
in the configuration.

To get back to the original question, yes, I think HSRP is worth keeping
around and shouldn't really have a line-item cost associated with it to the
customer.  I've worked with providers that charge an "HA" fee during
provisioning (and often a recurring one as well) for customers that want it,
but personally I think offering an HA network as a service provider should
almost be a given.

If you're still uncomfortable with the multiple vlans bound to one group
issue, there's also the 4948 model to consider.  It removes the issue of
having a million eggs in one basket at the customer aggregation level,
effectively has a 4000 series sup, and Cisco tested this out for us with
1500 HSRP instances running (lab documents available offline if you'd like
to see).  Alas, it does rise the aggregation costs a bit though.

Hope that helps,

Brad McConnell
CCIE #16147

-----Original Message-----
From: owner-nanog at merit.edu [mailto:owner-nanog at merit.edu] On Behalf Of
Randal Kohutek
Sent: Friday, May 11, 2007 2:21 PM
To: 'Mike Lyon'
Cc: nanog at merit.edu
Subject: RE: HSRP availability in datacenters?

I had read that on our original deployment, and it's a nightmare to keep the
documenation and configuration in synch. My personal opinion is that
potentially failing 16 VSIs over to the standby at once (because they're all
in the same group) - instead of just the affected ones - is poor policy.

I agree, 6500s or 4500s for distribution are where it's at ... Unfortunately
they cost a lot. Which is why the suits are considering financing them by
charging for the features they provide.

This has been a hot topic around the office, with all of us network guys
saying `keep hsrp everywhere` because it makes our phones ring less, but we
realize that network upgrades aren't free, which is making the non-IT folks
all antsy.

Regards,
Randal

> -----Original Message-----
> From: Mike Lyon [mailto:mike.lyon at gmail.com] 
> Sent: Friday, May 11, 2007 1:11 PM
> To: Randal Kohutek
> Cc: nanog at merit.edu
> Subject: Re: HSRP availability in datacenters?
> 
> Check out this article:
> 
> http://www.cisco.com/en/US/products/hw/switches/ps646/products
> _qanda_item09186a00801cb707.shtml#q1
> 
> Get rid of the 3550. Get youself a 6509 or 6513 :0
> 
> -Mike
> 
> 
> On 5/11/07, Randal Kohutek <nanog at data102.com> wrote:
> > We currently offer HSRP everywhere, the problem is that it doesn't 
> > scale on a budget. For example, a 3550 can do 16 HSRP 
> groups, limiting 
> > the number of customers that we can attach to (2x 3550s) to 
> 16. That's 
> > a lot of distribution infrastructure for 16 customers. Then 
> to scale 
> > that, say, to
> > 200+ customers, that means we have 12-13 pairs of distribution 
> > 200+ routers, each
> > with 2x gigE uplinks to the core ... Which means that 
> either (A) the 
> > core has to be really big or (b) we get fewer, more powerful 
> > distribution devices.
> >
> > This is where my employer is at now - I admit, we're tiny in the 
> > datacenter world - but the cost to aggregate 100+ HSRP 
> groups into the 
> > core, with room to grow, is pretty staggering for a smb.
> >
> > This why the suits are wondering if there is a revenue opportunity 
> > hiding somewhere to finance such a thing. Ah, the joys of 
> growing out 
> > of your britches :)
> >
> > Thanks for any continued response,
> > Randal
> >
> >
> >
> > > -----Original Message-----
> > > From: owner-nanog at merit.edu 
> [mailto:owner-nanog at merit.edu] On Behalf 
> > > Of Mike Lyon
> > > Sent: Friday, May 11, 2007 12:40 PM
> > > To: Randal Kohutek
> > > Cc: nanog at merit.edu
> > > Subject: Re: HSRP availability in datacenters?
> > >
> > >
> > > So is the question: you are selling transit to your customers and 
> > > you are wondering if you should charge your customer for allowing 
> > > them to use your HSRP gateway instead of a physical interface on 
> > > your router?
> > >
> > > Personally, if I saw a provider charging for that 
> service, I would 
> > > shy away from them. Only because it tells me they are 
> piece-mealing 
> > > their services and are cheap. I would think a good provider would 
> > > include that (and/or not sell it WITHOUT
> > > HSRP) in their sales offering. If for the only reason of customer 
> > > support nightmares. If you have your customers on HSRP 
> and you have 
> > > a router go down, you wont have them calling you every 
> five minutes 
> > > bitching at you...
> > >
> > > -Mike
> > >
> > >
> > > On 5/11/07, Randal Kohutek <nanog at data102.com> wrote:
> > > >
> > > > My cohorts in suits have begun wondering if HSRP is 
> standard for 
> > > > customer gateways, and from there wondering if it is
> > > something we should charge for.
> > > > I did some research and came up with mixed results; I'd
> > > like to hear
> > > > nanogers experiences with this:
> > > >
> > > > In your experience, do datacenters provide free HSRP
> > > gateways, or do
> > > > they make you pay for it?
> > > >
> > > >
> > > > Real world examples are better than Google :) Thanks, Randal
> > > >
> > > >
> > >
> >
> >
> 