This is an explaination of why some people mirror swap, and why some people have multiple swap partitions. Here are some definitions RAID1: Mirroring disks. The purpose is to always have two copies of your data. So **BY DEFINITION**, when you do a write, the system call will not return successfully until the data has been written to **BOTH DRIVES**. Not "one, and I'll do the other one later". Reads, on the other hand, can be taken from either disk, so you do get a pseudo-striping effect for reads. What is swap: Basically, swap==on-disk virtual memory. Any time you run out of RAM doing something, the OS tries to free memory, by either reusing memory that already exists on disk (eg copies of executables that already exist on disk, or shared libraries, etc) or writing a temporary copy of something in memory, to a swap partition. Multiple swap partitions: Solaris has the freedom to pick and choose which partition it will use, for any "swap out to disk" operation. Once it is on disk, it then has to do future reads [for that data] from that swap partition. It is not automatically mirrored. Questions and Answers Why put swap on a mirrored device? Note the definition of swap. If you are out of RAM, you start using disk space. If you have errors on that disk, you've basically had a memory failure. You're up the creek without a paddle. Why is swap-on-raid1 not as good as multiple swap partitions, for performance reasons? Because when a server is in a RAM shortage, you tend to get in a swap LOOP. Swap something in, play with it, write it out again. Repeat with some other section of virtual memory. So for essentially everything in swapspace, you will be writing to both disks at once. From a performance standpoint, you may as well only have one disk. What criteria does solaris use to balance across swap partitions? I dont remember. In theory, it could either do "this disk is busy, I'll use that one", or "I've already used this one lots, I'll use that one". Heck, it could do BOTH, I have no idea. Here's a URL with more great info on swap behaviour: http://www.sunworld.com/swol-04-1996/swol-04-perf.html Philip Brown -- phil@bolthole.com -- http://www.bolthole.com/solaris/