On Wednesday 05 December 2007 15:39:36 Carlos E. R. wrote: > The Wednesday 2007-12-05 at 06:48 -0700, Bill Anderson wrote: > >> For perfomance, yes, you are right. For safety, no, you are wrong. > > > > I don't see the safety issue as a major issue. The kernel avoids sending > > dirty pages to swap. Also, any time the application does a write, the > > dirty pages are sent to the buffer, and buffers aren't swapped. The > > kernel does not swap any kernel data structure. Having mirrored swap > > areas isn't going to protect buffers that kflushd hasn't sent to the > > disk. In this case, my opinion is that performance takes precedence. > > Performance takes precedence if the admin of that systems prefers > performance. If the admin prefers reliability (or needs), then reliability > takes precedence. > > If the machine is to run 24*7, perhaps with hot-swappable disks, then > reliability takes precedence - and in that case swap *must* go on raid. > > The trick is that with swap on raid, if one disk goes down the system > continues running till it is replaced with no impact on users or programs. > > If you don't, then applications will simply stop (in iowait for ever) or > the kernel will panic. That means downtime! It is a major issue. > > > This difference is documented. > > > -- > Cheers, > Carlos E. R.
If swap is a major issue you've clearly not got enough RAM ;) -- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
