>>>>> "jamal" == jamal <hadi@xxxxxxxxxx> writes:
jamal> I think we had this debate during your presentation ;-> Here are
jamal> my thoughts:
jamal> Bus Latency is not a problem as far as throughput is
jamal> concerned. This problem can be equated to *exactly* the high
jamal> RTT-BW problem in TCP. You just have to adjust your ring-buffering
jamal> accordingly. I dont think processing latency is an issue either;
jamal> even with your broken pcnet driver[1] you come up with a number of
jamal> 4007 cycles to process a packet. Get yourself a faster processor
jamal> ;-> So your assertion that "the 33Mhz, 32 bit PCI bus itself can
Further investigation revealed a logic problem that was actually processing
the packet to upper layers for that number, i.e. actually queuing the packet
to the BH.
I need to update the paper.
How much faster?
jamal> theoretically handle up to one and a half million (1428571 to be
jamal> exact) frames per second, or 50 10 Mb/s adaptors" is misleading.
jamal> I realize you say it is theoretical; however, ask people who use
jamal> Alexey's fast forwarding driver and they'll tell you they
jamal> definetly do more than 50Mbps.
I said 1,500,000 packets/s. I.e. GB ethernet. That isn't 50Mb/s.
I don't dispute 50Mb/s. I agree with it. I also agree that you can drop
packets at 150,000 packets/s. The stats say so.. it is a question of
optimization. The major problem that the PAX.ware 100 has is that the
receive ring must be kept on the PCI memory, which makes processing it very
slow.
We have considered using a DMA controller to copy it to system memory
in the background, but haven't done that yet. (We observe very high hits
for accessing the receive ring resident on card)
We were able to optomize things to get well below the 4007 reported
initially. A respin of card will integrate the classification co-processor
much more tightly to the MAC, so the receive ring will reside in system
memory, where one doesn't have to pay the PCI-tax to get it.
:!mcr!: | Solidum Systems Corporation, http://www.solidum.com
Michael Richardson |For a better connected world,where data flows faster<tm>
Personal: http://www.sandelman.ottawa.on.ca/People/Michael_Richardson/Bio.html
mailto:mcr@xxxxxxxxxxxxxxxxxxxxxx mailto:mcr@xxxxxxxxxxx
|