I agree with Vikas, that's a great article on VM swap file locations (i.e. *.vswp). However, in this case I'm pretty sure we're talking about the Guest OS swap (i.e. pagefile.sys on Windows, for example). Placing GOS swap on dedicated vmdk's (i.e. P:\ for pagefile) is really overkill and actually adds another point of failure, not to mention management overhead for the vmadmin.
I have never personally adopted this. Occasionally, this will come up in storage vendor health checks as a best practice. IMHO it's a cry for help and a last ditch attempt for them to buy time or explain why their array is experiencing [insert problem here]. That's never the problem, and I always push back if they try that on me. I agree it's good for dedupe on the non-paged volumes, but not worth it IMO.
It's not that I object to it as being viable, it would actually be pretty cool in some situations. For example, if you follow the best practice of sizing your GOS swap as 1.5 times RAM (still considered a MS best practice), then you get the advantage of having a dedicated logical drive letter inside the GOS where you can safely expand the disk size to account for increases in RAM size.
Further, dedicated page file vmdk's prevent low free space conditions inside the GOS system disk (especially on dynamically sized Windows pagefiles. Damn I hate those). I guess one final nice thing is the ability to easily see how bad things are crunching on those swap files... much easier on a dedicated datastore I would guess, and it prevents heavy paging (especially when misaligned) from become I/O bullies affecting other VMs.
Anyway, just a couple thoughts. I don't think either way is right or wrong, you can make arguments both ways; It really comes down to preference. Personally, I think defaults are good. The less moving parts the better.
PS - and damnit... not my intention but I may have said more nice things about separating them, but maybe I'm just being nice. I would never do it :smileysilly:
