Brandon wrote: > On Thu, Apr 10, 2008 at 12:59 AM, Nico Sabbi <[EMAIL PROTECTED]> wrote: > > yes, I limited the ARC only after having verified that the default > > setting (unspecified in /etc/system) was too slow. > > What would be a good value considering that the server has 2 GB > > of ram? > > Without knowing the specifics of your system, I can't begin to guess. > The ARC's upward limit is about 1/2 of physical memory,
According to the source, it's "3/4 of all memory, or all but 1G, whichever is more" (64-bit kernel only, 32-bit is different). http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/fs/zfs/arc.c#3231 3231 /* set min cache to 1/32 of all memory, or 64MB, whichever is more */ 3232 arc_c_min = MAX(arc_c / 4, 64<<20); 3233 /* set max to 3/4 of all memory, or all but 1GB, whichever is more */ 3234 if (arc_c * 8 >= 1<<30) 3235 arc_c_max = (arc_c * 8) - (1<<30); 3236 else 3237 arc_c_max = arc_c_min; 3238 arc_c_max = MAX(arc_c * 6, arc_c_max); > and it will > decrease as required by memory pressure from other > applications. > In theory, you should never have to adjust it. There are > some instances > where this is not the case, but for most scenarios > don't touch it. Hmm, in theory it should... A few days ago I tried to install snv_85 into a virtualbox guest, running on a 64-bit amd dual core box with 2GB of physical memory which is using snv_85. The snv_85 guest's virtual hdd was a file on a zfs filesystem. And since the new GUI installer was complaining that the guest's 768MB memory wasn't enough for a graphical installation, I increased the guest's memory allocation to 1GB. The result was heavy thrashing of the hdd during the attempt to install snv_85 into that 1GB virtualbox guest! I think it took over half an hour to stop that 1GB virtualbox guest through the GUI, until I got back control over the machine. I guess that was a case there the automatic arc cache tuning didn't work too well (didn't work at all). I completed the virtual box gui install test using a little less memory, IIRC I used 768+64 = 832 mbytes. And another case was an attempt to use "xsltproc(1)" on a big xml file, this time on an amd64 x2 machine with 4GB of memory, using zfs, and the xsltproc process had grown to use > 2GB of memory. Again heavy disk trashing, and I didn't had the impression that the arc cache did shrink enough to prevent that thrashing. This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
