Prev: DRAM Devices and Systems: A JEDEC-Sponsored Educational Event
Next: 100% free dell laptaps offerd by dell and Vodafone companies.
From: Morten Reistad on 30 Jul 2010 12:37 In article <rcl8i7-aco.ln1(a)ntp.tmsw.no>, Terje Mathisen <"terje.mathisen at tmsw.no"> wrote: >Bakul Shah wrote: >> On 7/29/10 12:57 AM, Terje Mathisen wrote: >>> (Skipping/delaying the checksum test would mean depending upon the end >>> node to detect the error.) >>> >>> BTW, is anyone doing this? Maybe in order to win benchmarketing tests? And bring down latency. A lot, actually. Naive store and forward of a 12400 bit (farming included) frame on a gigabit link uses around 86 us. Add some store/handling logic, and you are close to 100. Modern gigabit switches bring this latency down to below 10. and, yes, there are procedures to handle bad crc by signalling BAD at the end of a frame to the next hop. >> You can drop a bad CRC packet at a later point in the pipeline but before >> sending it out. > >I meant sending out _before_ you have received it, as soon as you have >the dest address. Yes, it is called "cut-through switching". Now go read the promo materiel about it. And, yes, it really helps on protocols like NFS. >Terje - mrr
From: Benny Amorsen on 30 Jul 2010 14:34
Morten Reistad writes: > Yes, it is called "cut-through switching". Now go read the promo materiel > about it. And, yes, it really helps on protocols like NFS. But is there anyone who does it for routing? And why did cut-through pretty much disappear for gigabit switching, only to reappear for 10 Gbps? Except for HP who seem to be stuck with store-and-forward, but maybe that is why they bought 3com. /Benny |