Re: What are the issues related to increasing Rx ring size ?
Bogdan Costescu <[email protected]> Tue, 5 Oct 2004 19:10:18 +0200 (CEST)
| Newsgroups | gmane.linux.drivers.vortex.devel |
|---|---|
| Message-ID | <Pine.LNX.4.44.0410051854190.3514-100000@kenzo.iwr.uni-heidelberg.de> |
[ Sorry for the late reply. I just came back from holiday... ] On Thu, 23 Sep 2004, Amitava Biswas wrote: > I want to use this driver under a situation when packets arrive in > bursts (few hundered to say one thousand small (64Byte) packets in > one burst). Maybe NAPI could also help. I think that I've seen NAPI patches for 3c59x posted on netdev. > Perhaps this simple information is available in the NIC specs, which > is not a public domain doc. I am expecting that some driver writer > has this, and can kindly share this information or the document > itself or a tip on how to get it from 3 com (I would be grateful). Sorry, but this is not possbile due to the rules that we agreed to when we downloaded the document from 3Com. There was somebody from 3Com that wrote some months ago promising to look after making the docs available again, but I last time I checked (~2 months ago) they were still not to be found on 3Com's site. > 2) What are the other issues which may be the hurdles or constraints > to the driver operation with large Rx ring size ? These may be both > - system / resource constraints and performance constraints. I would like to add to Donald's answer another issue: max_interrupt_work is a parameter that limits (among others) the amount of packets that can be processed in one call to the interrupt routine. The default values for max_interrupt_work and for the Rx ring size are pretty close (or identical, depending on the driver version) which allows a complete processing of a full Rx ring in one call of the interrupt routine. If latency is important, you should probably increase max_interrupt_work to be close to the Rx ring size - but be prepared for the system to spend lots of time dealing with network packets and not be responsive for anything else; otherwise, although the packets will be queued and not lost, they will not reach the upper layers until later on after the interrupt routine was called lots of times. -- Bogdan Costescu IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868 E-mail: [email protected] _______________________________________________ vortex mailing list [email protected] http://www.scyld.com/mailman/listinfo/vortex