On 12/8/06, Gerrit Renker <gerrit@xxxxxxxxxxxxxx> wrote:
I am almost fully convinced that it is necessary to armour the packet history list operations with read/write locks. The reasons: * list manipulation is asynchronous, packet_sent/send_packet fill the list, while packet_recv purges * send_packet may sleep so that meanwhile new entries are put on top * after applying patch 9 I got, for the first time, accurate RTT values [patch 9 puts all history work into packet_sent] It would be good to have further opinions/ideas on this.
I totally agree with this. I had looked into it at one point as I wondered whether this was part of our problems and at that stage I didn't as appeared to be OK. It might have changed since then or I might just have been wrong but none the less it is definitely worthwhile. Over time I'd like to see these move to ring buffers as locking around linked lists ruins concurrency, stalls etc but that is relatively minor. Ian -- Web: http://wand.net.nz/~iam4 Blog: http://imcdnzl.blogspot.com WAND Network Research Group - To unsubscribe from this list: send the line "unsubscribe dccp" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
