(Illustration by Gaich Muramatsu)
On Thu, Mar 10, 2005 at 02:12:42PM -0500, Daniel Savard wrote: > I tried to setup the server running the vice-setup script, everything > went fine until I reached the RVM data file. My disks system is LVM2 > managed. I created separated volumes for /vice, /vicepa, rvmlog and > rvmdata. All of them are logical volumes and /vice and /vicepa are > journalled filesystems (IBM JFS for Linux). The rvmlog is 75M and > rvmdata 1G. I have 1G RAM in this system and around 80G disks. > > Here is the message I am getting while the script is trying to setup > rvmdata: > > LOG file has been initialized! > > > Rdsinit will initialize data and log. > This takes a while. > rvm_initialize succeeded. > Going to initialize data file to zero, could take awhile. > ? Couldn't write to /dev/rootvg/rvmdata. > Your coda server is not completely setup. You will need > set it up by hand or fix the problems and rerun /usr/sbin/vice-setup. > > What is going on? Maybe we are trying to write 1024*1024*1024 bytes, while the partition has 1000*1000*1000 bytes. In any case, use 300MB RVM data size or something, not many people have used larger sizes all that much. We need a contiguous chunk of memory, and because there are whole areas already used by shared libraries, stack and reserved kernel memory we might already be getting a bit tight on memory with 1GB. Also, linux doesn't allow you to mmap a raw partition, so we have to allocate a big chunk of memory and copy everything in there during startup. If your RVM data is about the same or larger than your physical memory this will cause some significant swap activity since every page is considered modified. If you use a file as the backing store for RVM data, we simply mmap the thing and bits and pieces will be paged in and out as needed, only the modified pages will end up in swap. So this reduces memory/swap pressure and gives quite a noticable improvement in the startup time of a server. JanReceived on 2005-03-10 17:29:33