From owner-freebsd-net Wed Sep 13 17:16:16 2000 Delivered-To: freebsd-net@freebsd.org Received: from fw.wintelcom.net (ns1.wintelcom.net [209.1.153.20]) by hub.freebsd.org (Postfix) with ESMTP id 2E06F37B422 for ; Wed, 13 Sep 2000 17:16:14 -0700 (PDT) Received: (from bright@localhost) by fw.wintelcom.net (8.10.0/8.10.0) id e8E0GCX27001; Wed, 13 Sep 2000 17:16:12 -0700 (PDT) Date: Wed, 13 Sep 2000 17:16:12 -0700 From: Alfred Perlstein To: Bosko Milekic Cc: freebsd-net@FreeBSD.ORG Subject: Re: Clusters larger than PAGE_SIZE and contigmalloc() Message-ID: <20000913171611.E12231@fw.wintelcom.net> References: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2.4i In-Reply-To: ; from bmilekic@dsuper.net on Wed, Sep 13, 2000 at 08:13:05PM -0400 Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org * Bosko Milekic [000913 17:10] wrote: > > Hi, > > With the recent cleanups and SMP work that I'm involved with > revolving around mbufs and friends, I feel that it's about the time to > raise this issue. > > I'd like to know: > > Are there any people out there using the "large cluster" feature > (i.e. manually defining the cluster size, MCLBYTES, to be larger than > a PAGE_SIZE?) If so, how useful do you find this? Did you stumble across > any problems worth mentionning? Well I attempted to use clusters larger than PAGE_SIZE without contigmalloc not realizing that there was no bus_space_foo for mbufs. It wasn't fun debugging that. Mike Smith suggested that I investigate how NetBSD handles this situation, I looked and it seemed somewhat ok, but at a glance somewhat inneficient. > I'm wondering if it would be worth just scrapping this code, as > contigmalloc() doesn't help us in this case much anyway, and since most, > if not all of the code, that needs such a feature maintains its own free > lists and has its own allocator, which is somewhat more efficient as it > pre-allocates all of the required space while attaching (the last time I > checked). contigmalloc() may have trouble finding the required > contiguous physical pages after a certain period of uptime. I would > assume that when this was initially written that it was real nice to have > around, but I'm not sure if this is still the correct approach for > reaching our present goals. Well, if you could do the bus space stuff for mbufs that would be optimal, I'm sure the ethernet driver authors wouldn't have much trouble migrating over to it. -Alfred To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message