I really don't see why anybody in their right mind
-f0dder
Hey! You leave my mind out of this! It's already unstable enough as it is 
!-barney
- I've had the pagefile run away with me on some systems, not always identifiable as to why.
- I always limit the size of the pagefile ... call it quirky, but I do.
-barney
Humm, back when I used pagefiles, I used to set a largeish minimum bound, so that under normal operation and moderate load, the pagefile size would stay constant (and unfragmented!), whereas exceptional situations could grow as large as needed. That would still be my recommendation - but if you have badly written software that runs amok, putting an upper bound might be reasonable...
- Mindful of SSD stuff I've read recently, a RAMdisk pagefile reduces writes, although 'twould seem the jury is still out on that one.
-barney
Entirely disabling the pagefile entirely disables writes, though :-)
- I've had the pagefile on a USB stick - not fast, but no noticeable deterioration in performance. This is just another test.
-barney
Personally wouldn't do this, because of how flaky usb sticks are - using ReadyBoost might be a somewhat safer middle ground.
[Sidebar]
After I eliminated the pagefile, my memory usage decreased by seven (7) to ten (10) percent. Not certain just what that implies, but I wouldn't have know w/o the experimentation.
[/Sidebar]-barney
Memory management is
complex - some applications tune their usage depending on "free memory", and what "free memory" means is also up for debate. If you go by "unused physical RAM", you ignore the fact that filesystem cache can be quickly discarded to serve large memory requests.
[Addendum]If that memory is not otherwise being used, why not put it to use?[/Addendum]-barney
Indeed. It's worth noting, though, that memory used for a RAMdisk can't be use for normal filesystem cache - and if you allocate a large ramdisk but only use part of it, technically that's still unused memory. It's all about striking a balance :-)
I've got 16 gigs of memory in my current rig, but only have a 1 gig permanent ramdrive. That's large enough for (most) %TEMP%, firefox profile+cache, WebsiteWatcher profile+cache, and scratch space (it's usually ~50% full). Anything larger than that would be wasted memory - but for special occasions, I can quickly add a temporary disk.
Suffice it to say that the performance improvement was significant in terms of raising the WEI (Windows Experience Index) "Memory operations per second" subscore, but not the overall Base Score (which is determined by the lowest subscore).-IainB
Slight nag: putting
more RAM in a machine doesn't give you more memory operations per second, unless the RAM you're putting in is faster. (there's an exception if you go from single to two sticks because of dual-channel memory architecture, though). Unfortunately, for some reason the WEI score does seem to limit the score to 5.9 unless you have >4GB of memory (no matter how fast the RAM is) - which is plain misleading, IMHO.
As for disabling firefox disk cache, ho humm. Might be an advantage, but not having stuff cached between sessions? I chose the third way: moving firefox profile + cache to RAMdisk. Best of both worlds
Whilst I have so far made no objective measures of the effects of using ReadyBoost, I can report that it certainly seems to work as it should, and that it seems to provide some latency reduction, though I do not yet understand how to make the most of this. Presumably the max read/write speed of the USB RAM ("memory stick" or whatever you might call it) is a constraint, and so measuring that would seem to be useful.-IainB
Well, yes and no.
ReadyBoost only serves "random" I/O requests (i.e., smallish scattered-all-over reads), whereas large linear sequences are served from harddrive. It's done this way because normal flash memory (usb pendrives, SD cards etc., not full-on SSDs) is quite a bit slower than harddrives - I have a decent USB2 corsair pendrive that does ~20MB/s and ~24MB/s reads, whereas my Velociraptor does ~140MB/s. But once you start doing highly random 4kb I/O, those figures change dramatically - even a high-end HDD like the velociraptor drops in below 1MB/s, whereas flash ram holds up much better (contrary to common belief, they aren't 100% directly addressable, though, so there
will be some performance drop).
RB is somewhat intriguing, since it isn't just "more memory for the pagefile", it's actually a full disk caching layer - so unless you run an SSD, you might see some benefit from it even if you have large enough amounts of RAM (that would mostly be after a cold system boot, though, as memory filesystem cache would then get filled, and memory is insanely much faster than flash memory). I don't think I'd ever use RB, though - flash memory is too fragile, and enough ram + SuperFetch should be better.
Also note that ReadyBoost does both compression and AES128 encryption of the cache files, so there will be some CPU overhead.
Also, for those that might want it, there will probably be ways to make a backup HDD copy of files in a RAM disk/cache - if necessary - before terminating the process or before Shutdown. (Though I haven't played around with this yet.)-IainB
More than one RAM disk offers persistance, saving the memory content to disk on shutdown.
SoftPerfect RAM disk might be worth checking out
Recent Posts
- avoiding mysql is because of whOracle. It's a fork of mysql by the original mysql developer...
