Deaggregation Disease

-- Jared Mauch <jaredpuck.nether.net> wrote:

>On Fri, Jul 21, 2006 at 02:42:04PM +0200, Fredy Kuenzler
wrote:
>> 
>> Valdis.Kletnieksvt.edu schrieb:
>> >On Fri, 21 Jul 2006 09:51:33 +0200, Fredy
Kuenzler said:
>> >
>> >>>>Prefixes  Change          ASnum    
    AS Description
>> >>>>3263      0->3263         AS4151
       USDA-1 - USDA
>> >>>so I wonder what's wrong with them.
>> >
>> >I'm not sure which is more weird - a jump of
over 3K routes, or the
>> >fact that the starting point is zero....
>> 
>> Just to make it clear: AS4151 was 9 month ago. Now
we see history again 
>> with new actors. (I guess the actual increase was
done by various
ASN of 
>> RENATER).
>> 
>> I wonder why aggregating is that difficult.
>
>	It's not, people are just lazy and since "nobody
owns the internet
>man", or maybe "it's all a bunch of
tubes" there's nobody to force people
>to be good actors.  Perhaps it's time to bring back the
old /19
>filters that were started by sprint & such.
>

I was just thinking the same thing. 

- ferg


--
"Fergie", a.k.a. Paul Ferguson
 Engineering Architecture for the Internet
 fergdawg(at)netzero.net
 ferg's tech blog: http://fergdawg.blogspo
t.com/

On Fri, 21 Jul 2006, Fergie wrote:

>> 	It's not, people are just lazy and since
"nobody owns the internet
>> man", or maybe "it's all a bunch of
tubes" there's nobody to force people
>> to be good actors.  Perhaps it's time to bring
back the old /19
>> filters that were started by sprint & such.
>
> I was just thinking the same thing. 

As we push closer to the ipv4 route table limits of cisco's
6500/7600 
series (with anything less than Sup720-3bxl), I suspect lots
of networks 
are going to be forced to start doing some sort of filtering
of routes 
beyond just refusing >24-bit networks or cisco's going
to sell a lot more 
Sup720-3bxl's, FAN2 trays, and power supplies in the next
year or two.

------------------------------------------------------------
----------
  Jon Lewis                   |  I route
  Senior Network Engineer     |  therefore you are
  Atlantic Net                |
_________ http://www.lewis.org
/~jlewis/pgp for PGP public key_________

On Fri, 21 Jul 2006 13:59:35 EDT, Jon Lewis said:

> As we push closer to the ipv4 route table limits of
cisco's 6500/7600 
> series (with anything less than Sup720-3bxl), I suspect
lots of networks 
> are going to be forced to start doing some sort of
filtering of routes 
> beyond just refusing >24-bit networks or cisco's
going to sell a lot more 
> Sup720-3bxl's, FAN2 trays, and power supplies in the
next year or two.

The big question is, of course, whether to upgrade a 6500
and keep it on
life support, or bite the bullet and go for a whole new box.
 How much time
a -3bxl and careful filtering will buy you does depend
heavily on where in
the Internet you are - but I'm willing to bet that a good
number of sites
will go for the fork lift upgrade because there are *other*
pressing things
coming up that the 6500 won't do either.

Remember - it only takes *one* truly mission-critical
"must do" that a 6500
can't, and it's off to a less stressful corner of your
network for that long
slide into retirement (on the other hand, I'm sure in 2016,
there will *still*
be 6500's installed, just like I'm sure there's still
1996-vintage gear still
out there now...)

I'll concede that Jon is at least partially right -
*somebody* is going to
be selling gear... ;)

On Fri, 21 Jul 2006 Valdis.Kletnieksvt.edu wrote:

> The big question is, of course, whether to upgrade a
6500 and keep it on
> life support, or bite the bullet and go for a whole new
box.  How much time
> a -3bxl and careful filtering will buy you does depend
heavily on where in
> the Internet you are - but I'm willing to bet that a
good number of sites
> will go for the fork lift upgrade because there are
*other* pressing things
> coming up that the 6500 won't do either.

With a 3bxl, you won't need careful filtering.  All the
lower Sups top out 
at or slightly below 256k routes.  IIRC, the 3bxl claims to
support 1M 
ipv4 routes.  Anyone else care to guess at how far off 235k
routes is?

> I'll concede that Jon is at least partially right -
*somebody* is going to
> be selling gear... ;)

Yeah...I posted recently on cisco-nsp that I think cisco's
making a huge 
mistake not producing a Sup32-3bxl.  When the Sup2 can't
cope with "full 
routes" anymore, I suspect the Sup720-3bxl will
already have been 
obsoleted by some higher end Sup.  Then networks that would
have bought 
Sup32-3bxl's for the route capacity, and don't really need
the traffic 
capacity of the Sup720-3bxl will snap up Sup720-3bxl (and
the required 
fan2s and power supplies) off the used market while
bigger/richer networks 
upgrade to the Sup720-3bxl replacement.

------------------------------------------------------------
----------
  Jon Lewis                   |  I route
  Senior Network Engineer     |  therefore you are
  Atlantic Net                |
_________ http://www.lewis.org
/~jlewis/pgp for PGP public key_________

On Fri, Jul 21, 2006 at 01:59:35PM -0400, Jon Lewis wrote:
> 
> As we push closer to the ipv4 route table limits of
cisco's 6500/7600 
> series (with anything less than Sup720-3bxl), I suspect
lots of networks 
> are going to be forced to start doing some sort of
filtering of routes 
> beyond just refusing >24-bit networks or cisco's
going to sell a lot more 
> Sup720-3bxl's, FAN2 trays, and power supplies in the
next year or two.

It should be noted that the sup720-3a/3b tcam allocations
(cef 
maximum-routes) only gives 190k of the 256k theoretical max
to IPv6 routes 
by default. Anyone running a sup720 non-3bxl who has not
manually adjusted 
those cef maximum-routes is either blowing up or about to
blow up any day 
now, depending on how many internal routes they have and how
much 
filtering their upstreams are doing.

Of course this isn't a new problem, many of us are still
running old 
Foundry ironcore boxes with 700+ day uptimes and software so
old it came 
with 120k or 140k default maximum routes. Similiarly, cam
aggregation on 
such platforms (without enough cam to hold even close to
enough routes for 
a full table) is nothing new either. Cisco could easily
implement cam 
aggregation where they do not install a cef route entry if
there is a 
covering less-specific route pointing to the same
nexthop(s). It is 
hardly rocket science, and could extend the life of a 256k
route tcam 
platform for many years to come. But clearly Cisco would
rather just sell 
3bxl's. 

-- 
Richard A Steenbergen <rase-gerbil.net>       http://www.e-gerbil.net/r
as
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41
5ECA F8B1 2CBC)