Early on in this discussion engineers from multiple router vendors
stated publicly that that their engineers knew how to build routers that
would handle FIBs that were at least 4M entries or more (10M was
mentioned), using commodity hardware parts. It also became clear, at
least to a few of us talking on the side, that what the hardware guys
were telling one router vendor, in terms of what memory parts they will
have in a few years, is the same thing that they are telling other
router vendors. In various private conversations it also became clear to
me that the router hardware engineers working on the hardware forwarding
parts of routers were not even considering any major change to the
architecture, for the simple reason that they didn't need to.

There may be many reasons to consider making a major change to the way
that the Internet does routing and addressing, but the inability to
build equipment that can handle several million entries is NOT a valid
reason. 

Ross

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of
Steven Blake
Sent: 01 January 2009 11:47
To: Robin Whittle
Cc: RRG
Subject: Re: [rrg] BGP scaling limit?

On Thu, 2009-01-01 at 14:59 +1100, Robin Whittle wrote:

> Short version:  Limits on using PC hardware for router FIBs.
>                 Fast SRAM is needed, not DRAM.

[snip]

One can build a 4M+ prefix, 300M LPM/sec FIB engine (v4 or v6) using
commercially available FPGA + DDR3.

Regarding RIB processing: I'll accept the argument that rate*state
growth is exceeding Moore's Law.  However, most RPs I've seen are anemic
compared to commodity server hardware (for legitimate reasons).  So
there is plenty of room for improvement here.


Regards,

// Steve

_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg

Reply via email to