From mboxrd@z Thu Jan 1 00:00:00 1970 From: Peter P Waskiewicz Jr Subject: Re: ixgbe: [RFC] [PATCH] Fix return of invalid txq Date: Fri, 15 Jan 2010 01:00:20 -0800 Message-ID: <1263546020.2038.7.camel@localhost> References: <20100115053117.31513.82775.sendpatchset@krkumar2.in.ibm.com> <20100115.004456.15627093.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: "krkumar2@in.ibm.com" , "netdev@vger.kernel.org" , "Kirsher, Jeffrey T" To: David Miller Return-path: Received: from mga14.intel.com ([143.182.124.37]:8930 "EHLO mga14.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753773Ab0AOJAV (ORCPT ); Fri, 15 Jan 2010 04:00:21 -0500 In-Reply-To: <20100115.004456.15627093.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-ID: On Fri, 2010-01-15 at 00:44 -0800, David Miller wrote: > From: "Waskiewicz Jr, Peter P" > Date: Thu, 14 Jan 2010 23:58:17 -0800 > > > I've been trying to find time to add something like igb has, with a > > tiny Tx lookup table that maps CPUs into a smaller set of Tx queues. > > Why do you need "tables"? Just modulo the it, with whatever > optimizations you can come up with. > > Or do we not have enough data references in the TX path already? > :-/ > > I would suggest getting rid of the table in IGB too. > > Either "tables" are a good idea (I think they definitely are not) > or they are not. And whatever the decision is we should do it > consistently. net/core/dev.c doesn't use tables, it does the > subtraction modulo thing like Krishna does. What I've been thinking of is more for the NUMA allocations per port. If we have, say 2 sockets, 8 cores a piece, then we have 16 CPUs. If we assign a port to socket 0, I think the best use of resources is to allocate 8 Rx/Tx queues, one per core in that socket. If an application comes from the other socket, we can have a table to map the other 8 cores from that socket into the 8 queues, instead of piling them all into one of the Tx queues. Cheers, -PJ