Greetings,
I am researching possible replacements for our Internet edge routers, and wanted to see what people could recommend for a smaller chassis or fixed router that can handle current IPv4 routes and transition into IPv6. Currently we have Brocade NetIron 4802s pulling full IPv4 routes plus a default route. I've looked at Extreme, Brocade, Cisco, and a few others. Most range from 256k - 500k IPv4 and 4k - 16k IPv6 routes when CAM space is allocated for both. The only exception I've found so far is the Cisco ASR 1002, which can do 125k v6 along with 500k v4 routes at once. I'm curious if any other vendors have comparable products.
My concern is trying to find a router (within our budget) that has room for growth in the IPv6 routing space. When compared to the live table sizes that the CIDR report and routeviews show, some can't handle current routing tables, let alone years of growth. BGP tweaks may keep us going but I can't see how 16k or fewer IPv6 routes on a router is going to be viable a few years from now.
Thank you,
Chris Enger
From: Chris Enger
Sent: Tuesday, March 08, 2011 4:15 PM
To: 'nanog@nanog.org'
Subject: Internet Edge Router replacement - IPv6 route table
sizeconsiderations
Greetings,
I am researching possible replacements for our Internet edge
routers, and wanted to see what people could recommend for a smaller
chassis or fixed router that can handle current IPv4 routes and
transition into IPv6. Currently we have Brocade NetIron 4802s pulling
full IPv4 routes plus a default route. I've looked at Extreme,
Brocade, Cisco, and a few others. Most range from 256k - 500k IPv4
and
4k - 16k IPv6 routes when CAM space is allocated for both. The only
exception I've found so far is the Cisco ASR 1002, which can do 125k
v6
along with 500k v4 routes at once. I'm curious if any other vendors
have comparable products.
The NetIron XMR will get you 1,000,000 routes.
I've been very happy with the Juniper J4350/6350 series.
Owen
1M of what kind of routes?
A Sup720-3bxl can do 1M v4 routes or some split of v4 and v6 routes. i.e.
L3 Forwarding Resources
FIB TCAM usage: Total Used %Used
72 bits (IPv4, MPLS, EoM) 622592 348493 56%
144 bits (IP mcast, IPv6) 212992 263 1%
That's 1M IPv4 routes, IIRC. Put IPv6 into the mix and that 1M quickly
shrinks.
Frank
That's 1M IPv4 routes, IIRC. Put IPv6 into the mix and that 1M quickly
shrinks.
<snip>
I am researching possible replacements for our Internet edge
routers, and wanted to see what people could recommend for a smaller
chassis or fixed router that can handle current IPv4 routes and
transition into IPv6. Currently we have Brocade NetIron 4802s pulling
full IPv4 routes plus a default route. I've looked at Extreme,
Brocade, Cisco, and a few others. Most range from 256k - 500k IPv4
and
4k - 16k IPv6 routes when CAM space is allocated for both. The only
exception I've found so far is the Cisco ASR 1002, which can do 125k
v6
along with 500k v4 routes at once. I'm curious if any other vendors
have comparable products.
Mx80 Doesn't use CAM for route lookups and the actual number of routes
is 72MB of RLDRAM holds will vary based on the distribution of sizes but
it's north of 2 Million ipv4 routes.
From: Frank Bulk [mailto:frnkblk@iname.com]
Sent: Tuesday, March 08, 2011 10:13 PM
To: George Bonser; nanog@nanog.org
Subject: RE: Internet Edge Router replacement - IPv6 route table
sizeconsiderations
That's 1M IPv4 routes, IIRC. Put IPv6 into the mix and that 1M
quickly
shrinks.
Frank
Yes, you have two choices with the Brocade gear. You can have it set to
install the /64 prefix in the table making a v6 route 2x as expensive as
a v4 route (default behavior) or you can have it use the entire 128 bit
destination where it becomes 4x more expensive.
The ipv4-ipv6-2 CAM profile in 5.1 gives 768K v4 routes and 64k v6
routes which should be good for quite a while. That is provided you
aren't using MPLS VPNs. If you are, the best you can get using static
CAM profiles is multi-service-2 which gives you 384K v4 and 128K v6.
But you can put the thing in dynamic cam mode where unused entries can
be aged out saving resources for routes you never really talk to:
Dynamic mode - In the dynamic mode, routes are entered into the CAM
dynamically using a
flow-based scheme, where routes are only added to the CAM as they are
required. Once routes
are added to the CAM, they can be aged-out when they are not in use.
Because this mode
conserves CAM, it is useful for situations where CAM resources are
stressed or limited.
Static mode - In the static mode, routes are entered into the CAM
whenever they are
discovered. Routes are not aged once routes are added to the CAM and can
be aged-out when
they are not in use.
The default behavior is static mode. I use dynamic mode on dual-stack
gear at the moment with the ipv4-ipv6-2 CAM profile.
Chris,
With address exhaustion and deaggregation, the table is only going to
get bigger so choosing anything now that can only handle anything
south of 1M routes is not a wise investment.
Several posters have recommended ASR1002 and MX80. I use both of these
platforms in my environment and have been quite pleased with both.
ARA100x. Cisco has lower/cheaper options here including a 1RU device.
I don't have the specs handy, but these are lacking in scalability
that you will most likely need. I believe the forwarding cap is 2.5G.
With the ASR1002, you can start up with the 5G forwarding board.
The MX80. There are several models/bundles. A good choice for you may
be the MX80-5G. Incidentally, the "5G" does not mean 5gig. It ships
with a 20 port ge MIC that will do line rate. The other MIC and the
on-board 4X 10GE are disabled. As previously mentioned, it doesn't use
TCAM so your V4, V6 routes don't share finite resources with each
other or MAC entires, etc. If you're familiar with the benefits if
JUNOS - once you've used it for awhile - it's hard to go back.
If your environment is rapidly growing, stay away from low CAM
limits,anything that's runs in software, (C7200, C7330, J6350), and
make the jump to line-rate hardware devices.
-b
Thank you everyone for the suggestions both on and off list. We will be looking at a few additional devices along with what we have researched.
Thanks,
Chris
My concern is trying to find a router (within our budget) that has room for growth in the IPv6 routing space. When compared to the live table sizes that the CIDR report and routeviews show, some can't handle current routing tables, let alone years of growth. BGP tweaks may keep us going but I can't see how 16k or fewer IPv6 routes on a router is going to be viable a few years from now.
Thank you,
Chris Enger
Does anyone think that the IP6 routes will grow like IP4 routes have?
With most organizations being granted IP6 /32's - e.g. something larger
than they could ever use - wouldn't you expect the number of routes to
be much much fewer than with today's IP4 setup where even small
organizations often have multiple routes, and big organizations may have
hundreds?
When sizing routers, shouldn't we be looking at the number of expected
ISPs (AS's) active on the Internet, within the anticipated lifetime of
the router? If so, then the question becomes how many is that - 16k
seems very shortsighted, 128k maybe overkill (at least, for now). We're
currently at 37k AS's (CIDR Report). So 64k IP6
routes would probably be the minimum that I would accept on a new single
homed router. If I expected to act as a carrier, or participate in equal
cost BGP routing on a multi-homed router, I'd need more.
As IP6 adoption grows, and networks start to de-aggregate, 128K IP6
routes sound like a better number for the second or third revision of
"IP6 ready" gear that would be purchased in 5+ years.
--Blake
My concern is trying to find a router (within our budget) that has room for growth in the IPv6 routing space. When compared to the live table sizes that the CIDR report and routeviews show, some can't handle current routing tables, let alone years of growth. BGP tweaks may keep us going but I can't see how 16k or fewer IPv6 routes on a router is going to be viable a few years from now.
Thank you,
Chris Enger
Does anyone think that the IP6 routes will grow like IP4 routes have?
With most organizations being granted IP6 /32's - e.g. something larger
than they could ever use - wouldn't you expect the number of routes to
be much much fewer than with today's IP4 setup where even small
organizations often have multiple routes, and big organizations may have
hundreds?
Most end user organizations should be receiving a /48 per end site.
Most ISPs should probably get something larger than a /32 unless they
have fewer than 48,000 customers or so.
The number of prefixes per ASN in IPv6 should probably end up around 1.75.
The current average in IPv4 is approximately 10.
When sizing routers, shouldn't we be looking at the number of expected
ISPs (AS's) active on the Internet, within the anticipated lifetime of
the router? If so, then the question becomes how many is that - 16k
seems very shortsighted, 128k maybe overkill (at least, for now). We're
currently at 37k AS's (CIDR Report). So 64k IP6
routes would probably be the minimum that I would accept on a new single
homed router. If I expected to act as a carrier, or participate in equal
cost BGP routing on a multi-homed router, I'd need more.
16k is extraordinarily short-sighted. It will be at least 30,000 just to
duplicate the current IPv4 network. 128k is probably reasonable
headroom. I wouldn't want to go any smaller, given the likelihood
of some multiple slightly greater than 1 prefix per AS.
As IP6 adoption grows, and networks start to de-aggregate, 128K IP6
routes sound like a better number for the second or third revision of
"IP6 ready" gear that would be purchased in 5+ years.
I would say 128k is more like the minimum for anything I would buy
now.
Owen