Date: Tue, 13 Feb 1996 11:06:49 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: T0682740@pmail.tepco.co.jp Cc: hackers@FreeBSD.org Subject: Re: VM question Message-ID: <199602131806.LAA23438@phaeton.artisoft.com> In-Reply-To: <3120A0C8.2A46.001@pmail.tepco.co.jp> from "T0682740@pmail.tepco.co.jp" at Feb 13, 96 11:31:36 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> I posted the following mail to questoins@freebsd.org but no one > answered unfortunately. So please let me post it here again with one > additional question. I know that FreeBSD adopted Mach VM system. Does > this mean FreeBSD can have an external pager as Mach can? > > --------- > Hi, > > chuck@fang.cs.sunyit.edu said: > > I was going through the tutorials that were presented at USENIX > > last month and found a paper (from the "An Introduction to UNIX Kernel > > Internals" tutorial) called "A New Virtual Memory Implementation for > > Berkeley UNIX". > > I'm quite interested in this paper. How can I get it? Is it located > somewhere on the Internet? > > Also, according to the latest snapshopt announcement, it has a > VM performance enhancement over the 2.1.0-RELEASE. Could someone > briefly explain about it? Thank you in advance! > > --------- The Usenix papers are generally available for anonymouse FTP, both by the authors home site, and from the Usenix SAGE group ftp site. The SAGE FTP site is: ftp://ftp/sage.usenix.org/ FreeBSD does not have a mach VM system any more. The MACH VM system does not operate (implicitly) on a unified buffer cache scheme. Cache unification is an important win. External pagers cause crossing of protection domains. The use of external pagers is one of the reasons MACH is so slow; for a full list of speed problems in MACH, you should check out the Chorus papers. Chorus is a competing microkernel, and is the basis of much work at USL over the past 2 years or so. You could implement an external demand pager on the basis of trapping the SIGSEGV signal, and implementing mapping of resources as a result; you would robably be better off providing a kernel interface, however, since you would have to coalesce domains to keep the number of entries small enough to avoid problems. There are a couple of Usenix papers on this technique as well; other papers are available at the CS department of Washington State University's FTP server. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199602131806.LAA23438>