From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dave Hansen Date: Tue, 27 Sep 2005 14:25:17 +0000 Subject: Re: [Lhms-devel] Re: [PATCH 0/4] V4 ia64 SPARSEMEM Message-Id: <1127831117.10315.91.camel@localhost> List-Id: References: <1127779977.10315.6.camel@localhost> In-Reply-To: <1127779977.10315.6.camel@localhost> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: linux-ia64@vger.kernel.org On Tue, 2005-09-27 at 07:45 -0500, Jack Steiner wrote: > I haven't looked in detail at SPARSEMEM recently - is one config > file for ALL hardware platforms easy to do without significantly > compromising performance of any individual platform? On the ppc64 and i386 machines where I've done testing, it hasn't been an issue. Usually, the NUMA-Q is a nice canary and will fall over dead if we're doing any large amount of crazy, new, cross-node accesses. Compared to discontigmem, it actually gets a wee bit (~1-2%) faster on thinks like kernbench and SDET. But, we don't have a vmem_map on i386, so it's not an apples to apples comparison. I'd love to see some head-to-head performance numbers, though. -- Dave