Thanks, I've got a number of replies regarding, overall, the possible fault in 
the basis of my question - that is, that letting a system get into swapping is 
bad, so if that happens - the tasks or hardware should be reevaluated. And also 
that SSDs wear out fast so using them for swap is a bad idea in general.

While I agree with the validity of these points, they do not quite reply to the 
original question.

So let me generalize it a bit:
* There are some cases where swap/paging is needed and changing the hardware 
(RAM) or the tasks is not viable for whatever reason the system's architect or 
user may think up. Anyway - open OSes are about providing a way to make choices 
and tradeoffs, no?
* Some devices are better suited for use as swap areas, and being better in 
IOPS - they are also usually more expensive per GB and may be prohibitively 
small for a given swap requirement. If SSDs are a bad example, think 15kRPM 
disks vs. 5400RPM :) Or a dedicated small disk for swap vs. larger disk shared 
with other data-storage (and access) tasks.
* Sometimes extra swap might be needed as an "emergency safety net" when you 
expect your system to have some workload that is out of the ordinary, maybe 
once in a lifetime, which would require lots of VM and your spec'ed RAM and 
fast swap for its normal workloads are not going to be adequate - but just for 
a few days. So you add a cheap drive or a USB stick or whatever as a 
precaution, but you don't want it to be used for every other paging operation.

It seems to me that the Linux OS had considered these arguments long ago and 
added the concept of swap priority: AFAIK, the lowest priority-numbered areas 
are round-robined first; when these devices fill up, higher numbers are 
round-robined, etc.

It also seems to me that Solaris had no such implementation or concept.

So my questions are:
1) Is it true (that there are no swap priorities in Solaris as of now)?
2) Are there other sysadmins who think these might be useful in their own 
practice? 
Are any of you using this with your Linux boxes? Yes/no - why? ;)
3) Are there any developers who can estimate how easy or hard it is to 
implement, including standardization with the ARC (modelling after Linux, it 
might be an additional option to "swap -a" and a "mount option" in 
/etc/vfstab)? :)

//Jim
-- 
This message posted from opensolaris.org
_______________________________________________
sysadmin-discuss mailing list
sysadmin-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/sysadmin-discuss

Reply via email to