(Illustration by Gaich Muramatsu)
On Wed, Dec 01, 1999 at 05:51:10PM -0800, Phil Nelson wrote: > > >What platform? Solaris, Linux, NetBSD? Did you map more RVM data than > >physically available memory? > > Sorry I didn't answer all your questions ... > > a) NetBSD 1.4.1 > b) real mem 16M > c) RVM data 130M Phil wrote earlier: > (RVM data size, 266532, 133Mhz pentium ...) But with 16M of memory you would have brought a linux-2.0 box to it's knees. NetBSD doesn't suffer (as much?) from the double caching. Once we hit the swap on linux 2.0 (possibly even 2.2), there are copies of the same data floating around in the buffercache, the pagecache, and the swapcache. Or something like that. That slows a server down to a crawl. I just rebooted by laptop into 16MB, and started a server. Linux-2.2.13, Pentium 166, 16MB memory, 130MB rvm data (anonymous mapping) 21:09:30 Resource limit on data size are set to 2147483647 ... 21:09:30 Setting Rvm Truncate threshold to 5. (start of mapping) 21:10:50 Partition /vicepa:... (end of mapping) 21:11:07 File Server started ... Total startup time about 97 seconds, and my processor is faster than yours... I almost can't wait to try this with the private mappings. > USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND > root 2118 0.0 13.8 3624 2200 ?? S<s 3:19PM 0:15.61 codasrv -mappr Ok, this doesn't seem to account for the 130 megs of memory mapped RVM in the process. I've got: USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND root 534 2.9 64.8 121572 9604 ? S < 21:09 0:07 codasrv -sa 4 -t 1000... > >Did the rvm_basher turn up any more problems? > Just that one ... and I haven't figured out the problem. It happens > with anonymous and private mappings. I'll look into it furthur. It looked to me like it was trying to commit a larger transaction than could be fitted in the log. And failing on anonymous mappings could be an indication towards a bad parameter somewhere. JanReceived on 1999-12-01 21:34:30