A call for the future. Was: Re: Verio Decides what parts ofthe internet to drop

Randy Bush randy at psg.com
Wed Dec 8 23:09:25 UTC 1999

>>> Full route table size is not a problem. You can burn a hard disk as you
>>> mentioned to store it.  The issue is getting data in and out of the
>>> processor, i.e. number of pins.  Core flows are not ameneable to caching.
>>> This approach will fail the first time you see a new packet and need to
>>> swap from hard disk.
>> Not that it would be very economical, but what are the technical
>> implications of using a solid state device (such as the Quantum's
>> RUSHMORE NTE series) instead of a normal hard drive?
> Interesting question... even though it's significantly faster than an hard
> drive, it does have some inherent bottlenecks such as a maximum number of
> operations per second which might be a little stifling on a backbone core
> router :) Still, I've never actually tried *that*, so don't know for sure.
> There's also latency in other areas - that leads me to think that regular
> memory is still faster. Finally, it would be attached to the host system
> through a bus (SCSI, whatever) that's a lot slower than the internal
> memory bus.
> These kinds of devices tend to be a better fit for systems that doesn't
> have extreme time limitations on processing data such as for mail spool
> files, etc.

reality check:

if you had 100TB of on-ASIC SRAM you would still be screwed.  you can't
afford the PER-PACKET LATENCY of telco number style portability REFERRAL.

once again: ip is a connectionless protocol.  each packet is potentially a
new route.

telcos don't mind a second or two in call setup, because it is CALL SETUP,
not 42 times a second.

[ credit scott bradner for making this quite clear even to me in some ietf
bar ]


More information about the NANOG mailing list