From owner-freebsd-hackers Wed Jan 19 10:49:57 2000 Delivered-To: freebsd-hackers@freebsd.org Received: from apollo.backplane.com (apollo.backplane.com [216.240.41.2]) by hub.freebsd.org (Postfix) with ESMTP id 5FF07152D7 for <freebsd-hackers@FreeBSD.ORG>; Wed, 19 Jan 2000 10:49:55 -0800 (PST) (envelope-from dillon@apollo.backplane.com) Received: (from dillon@localhost) by apollo.backplane.com (8.9.3/8.9.1) id KAA39586; Wed, 19 Jan 2000 10:49:52 -0800 (PST) (envelope-from dillon) Date: Wed, 19 Jan 2000 10:49:52 -0800 (PST) From: Matthew Dillon <dillon@apollo.backplane.com> Message-Id: <200001191849.KAA39586@apollo.backplane.com> To: Jonas Bulow <jonas.bulow@parallelconsulting.com> Cc: FreeBSD Hackers <freebsd-hackers@FreeBSD.ORG> Subject: Re: UVM vs FreeBSD VM system References: <3885CDCD.E033801@parallelconsulting.com> Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG :Hi, : :How does the UVM system compare to the VM system in FreeBSD? Are there :any benchmark tests or research results in this area? Well, UVM is a much better design then the *original* Mach VM subsystem in 4.4. FreeBSD, however, does not use the original Mach VM subsystem anymore (thanks to John Dyson). UVM and FreeBSD's VM subsystem should have roughly similar performance. http://www.ccrc.wustl.edu/pub/chuck/tech/uvm/ http://www.daemonnews.org/200001/freebsd_vm.html The main focus of UVM was to deal with Mach's deep-stacking of VM Objects. They replaced the traditional shadow chains with a two-layer system. FreeBSD wound up keeping the shadow chains but adding an optimization that tended to restrict their depth. See the "All Shadowed" case in my paper on daemonnews. UVM has a lot of cool ideas especially in regards to avoiding program stalls when the system issues write-I/O. However, it also has some problems. I think FreeBSD handles VM/Filesystem coherency better at the moment (as of USENIX NetBSD/UVM did not have a coherent VM/Filesystem cache). UVM incorporated a number of FreeBSD ideas such as page clustering. FreeBSD is not without problems either. FreeBSD's VM system has more stall points then UVM's. In anycase, the UVM paper listed at the URL contains a good description of UVM and also compares it (roughly) with other operating systems, including FreeBSD. -Matt Matthew Dillon <dillon@backplane.com> To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message