Date: Thu, 1 Jun 2000 22:48:59 -0700 From: Arun Sharma <adsharma@sharmas.dhs.org> To: Mark Blackman <tmb@sophos.com> Cc: arch@FreeBSD.ORG Subject: NUMA (Was Re: Preemptive kernel on older X86 hardware) Message-ID: <20000601224859.A17524@sharmas.dhs.org> In-Reply-To: <20000531172432.A7547@sophos.com>; from Mark Blackman on Wed, May 31, 2000 at 05:24:32PM %2B0100 References: <200005301715.LAA27213@berserker.bsdi.com> <20000531172432.A7547@sophos.com>
index | next in thread | previous in thread | raw e-mail
On Wed, May 31, 2000 at 05:24:32PM +0100, Mark Blackman wrote: > On a nearly unrelated note, anybody care to venture > how tough would it be to modify *BSD* to run on a NUMA system > instead of an SMP solution. Most of the SMP optimizations (per CPU data structures to reduce locking) will help NUMA too. Additionally, (a) malloc(9) needs to be rewritten to allocate memory on the local node. (b) page allocation algorithms (vm_page_alloc) need to be rewritten. (c) some amount of process migration and page migration need to be implemented. (d) primitive load balancing (e) APIs to provide hints to the kernel on how to allocate shared memory. (f) kernel text and read only data replication. (g) Page cache replication for read only pages (feasible ?) -Arun To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-arch" in the body of the messagehelp
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000601224859.A17524>