"routing table slots" and the real problem
Mr. Dana Hudes
dhudes at graphnet.com
Mon Mar 3 15:13:15 UTC 1997
I hear you but the problem in 'open systems' is not memory or cpu speed:
it's the backplane.
I worked on the end of the NSFnet NSS project and with it forward to
where we had later versions
of software (including gated) and new cards intended for the 6611 but
with software rewritten
by clever people and fixes to the AIX kernal by a very smart programmer.
We had port density
-- 8 sync ports on a card, 2 sync + LAN (ether or tokenring your choice)
-- and we had high
end cards (FDDI and T3 from the NSFnet days, ATM over T3 and over TAXI
at the very end).
But an Rs/6000 is not a gigabit router. The theoretical max on the
backplane was 622Mbit and that
includes IP forwarding and disk I/O. I assure we did not hit 620Mbit
even going T3 <->T3.
Then there is the small point that an RS/6K is not exactly cheap like a
PC.
I have no idea whatsoever of the throughput capacity of PCI bus on a PC
but everyone in the
gigabit game is using switched backplanes rather than a shared bus. The
nature of the switch
is different in different vendors of course (Netstar is a 16way
crosspoint switch, for example,
while I heard someone was building a router with an ATM OC3 backplane).
The issue is not only whether open systems can take full routing and do
the route computations
because I'm sure they can (the Route Arbiter does). The question is
whether they can do that AND forward packets in today's (multi)gigabit
core on a normal shared bus architecture.
Noone I know is selling ATM backplanes for PC's .....
Dana
Joseph T. Klein wrote:
Warning -- I feel a diatribe emerging. ;-)
Afordability is primarily a question of how large your existing
base of legacy routers is and your cash flow.
You can build a box using a free versionof Unix (FreeBSD, NetBSD,
Linux, or whatever your religion of the day), off the shelf
hardware, and gated to route a full backbone routing table
(memory and CPU are cheap) for less than $3K. This is less
than the cost of a single interface card for a Cisco 7xxx!
Kids; do not try this at your gateway without adult supervision.
:-)
We are re-designing the Internet to make up for the fact the
largest manufacturers of routers has been slow to develop and
deploy systems that can keep up with the growth curve. A lot of
this comes down to size of the memory bus on low cost systems.
If port density was not so poor on general purpose hardware, we
would have been far better off deploying "open systems" for routers
rather than what exists today.
I have always liked my Ciscos, but I truly love routing with my Unix
systems running gated. ... now If I could find some cheap
channelized
DS-3 cards for a DEC AlphaStation 500. ;-)
I may be talking out of my hat here, but I suspect a DEC
AlphaStation
500 with 256M of RAM ranks pretty well against a 75xx.
Somebody, dig up the stats for me ...
If router manufacturers worked on hardware and all used an open
software standard ... such as gated ... we would all be better off.
Open standards allow all of us to benefit from the work of others.
The old Unix Guru's mantra is 'build on the works of others.'
Let us not make the mistakes of the 1890s and associate domination
of the market by oligopolies as good capitalism. Big corporations,
like big government, tend to move slowly.
Open markets NOT dominated by a single large player is GOOD
capitalism. It increases the pace of innovation and prevents
price fixing. It makes for a healthy, dynamic, marketplace.
This holds true for routers, backbone providers, toasters
and operating systems (sorry Bill)
open standards = open markets
Open standards prevent the failures of a single market
player from inhibiting the growth of the industry.
Open standards lower the cost to upgrade large installed
systems.
Reductions in the federal budget are squeezing R&D expenditures in
the US to an all time low. Large corporate downsizing and corporate
mergers have done the same for most large corporations. The bulk
of innovation in the US will come from small companies and
development consortiums.
It is from these that the next generations of routers will
come. Open standards make the rapid utilization of new
technologies possible and fuel the growth of small companies.
The Internet is a great place for consorting on standards.
This is what is cool about the IETF!
Standards do not keep the big boys from playing ...
Cisco and Bay could easily join in an open standard
for router software. It would not be hard to have interoperability
between the IP portions of IOS and gated.
IOS is the PL-1 of routers. Bay's management reminds me of CICS. ;-)
Back to the subject ...
You CAN also use the RA (where available) to reduce your routing
overhead, save memory and reduce CPU usage. (The RA runs a hacked
version of gated that calculates large routing tables quite well.)
Hmm ...
router $100,000 amortized over 3 years = 2,800/month
going DS-3 price at a NAP with line = 7,000/month
engineer $70,000 per year min. = 5,900/month
overhead for a small company = 20,000/month
$50+/month/mile for OC-3 lines ... don't even talk about
local loop costs!
Routers connect customers.
customers = cash flow.
The highest cost of running a national network is not buying
routers,
it is bandwidth, staff, and administrative overhead.
Router cost is primarily a factor for smaller networks with limited
cash flow.
I contend ...
It is the ISPs who try to be dual homed with 'routing tricks' rather
than using edge routers that can process a core routing table, who
contribute most to routing instability.
Boardwatch stated that 14% of ISPs are dual homed. I would bet
that 70% of those do not use routers capable of processing a core
routing table.
Anybody have any stats?
We need cheap routers that run BGP4 and can eat a core routing
table.
2501s just don't hack it in dual homed configurations ... and most
small guys just don't wish to blow $50,000 on putting 7505s at the
edges of their networks.
--- On Sun, 02 Mar 1997 13:48:46 -0500 Paul Ferguson
<pferguso at cisco.com> wrote:
> At 01:39 PM 3/2/97 -0500, Perry E. Metzger wrote:
>
> >
> >True enough. Of course, this doesn't mean that we can't have
routing
> >table growth, as we will have processor capacity growth, but it
does
> >mean that the growth of the routing tables must be kept in line
with
> >what the router processors can do.
> >
>
> True enough. However, it might also be novel to keep the cost
> down to a level that people can actually afford.
>
> - paul
>
---------------End of Original Message-----------------
--
From: Joseph T. Klein, Titania Corporation http://www.titania.net
E-mail: jtk at titania.net Sent: 13:25:09 CST/CDT 03/02/97
"They that can give up essential liberty to obtain a little
temporary
safety deserve neither liberty nor safety."
-- Benjamin Franklin, 1759
More information about the NANOG
mailing list