On Tue, 2006-01-31 at 16:48 -0800, David S. Miller wrote: > Why? >
We have an OEM customer who finds the current rx limit of 255 descriptors too low. IIRC, they are doing some distribution of real-time data and occasionally they get a big burst of incoming 300-byte packets. > To me, anything larger than 512 RX descriptors is overkill. > It's a buffer, just like the queue of receive sockets for > listening connections, and if you can't keep up then either: > > 1) Something is misdesigned or behaving badly > 2) You lack the compute power to process the packet load > > And in either case the best thing to do is drop frames and > not try to paper around the problem with huge RX descriptor > counts. Agreed in general. That's why the default is still 100 with one ring of 255 descriptors. Most users will not change this default setting. > > The complexity resulting from this expansion is yet another > argument against this. It's really not that complicated. It just allows additional rings to be added when the rx ring size is higher than 255. > > And with Van Jacobson net channels, none of this is going to > matter and 512 is going to be your limit whether you like it > or not. So this short term complexity gain is doubly not > justified. > > So, I definitely NAK this patch. > - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
