PrimoCache Settings for Best Performance

[Pliqui]

Hello All,

I recently bought PrimoCache after the trial. Helped me a lot with Vmware. Before PrimoCache, when i had some vms up, and tried to copy something all my pc gets really slow.

I want to optimize my settings and perhaps more experience people with PrimoCache can point me at the right direction. Note that i play a lot too.

First my machine Specs:

CPU: Core I7 2600 3.8Ghz
RAM: 1600 32GB G Skill Ripjaws
HDD0: 2 x Crucial M500 240 SSD Raid0 128KB Stripe Size (C: Drive for boot + games + some apps)
HDD1: 2 X WD Caviar Black 1TB 7200rpm 128KB Stripe Size (D: Apps + non ssd games + boot drive from all my Vmware workstation machines)
HDD2: 2 x WD Caviar Green 3TB 5200rpm (F: and G: movies, series, backups, extra storage for VM)

All my drives are formatted in a 4k blocks (windows default). For starters should I match the blocks of primo vs hdds?

NOTES:

.- Drives F and G are not important to be in the cache, what really matters are C and D.
.- Tonight was testing and cache was reset. Also my cache never goes up of 20%-30%, only when i was installing and patching a windows 7 and 8 to download all the patches to my update accelerator proxy, cache went really high for me 50%-55%

C Drive Raid0 SSD


D Drive Raid0 WD 1TB 7200rpm


Thanks in advance and really appreciate your help here

[InquiringMind]

Bear in mind that Windows does its own caching of disk data at file level, so Primo's level 1 cache will only get a look in if this cache doesn't contain the data requested. This is the most likely reason for your 20-30% cache usage figure (mine tends to be the same unless I spend lots of time on a single game, when it may go up to 40-60%).

Given Windows' file cache can grow to multiple GBs in size and will give up memory if needed by another process, it may not be worth using Primo's L1 caching unless you can allocate significantly more memory than Windows' cache uses (Task Manager should report the current/largest cache size). On the other hand, Primo L2 will be of no use on C: and running it on D: would require you to allocate one of your SSDs, breaking your RAID setup.

Whether PrimoCache would benefit your setup/usage comes down to the OS you run. If it is 32-bit, then a definite yes as Primo L1 can use the memory over 4GB that would otherwise be ignored - you could allocate 28GB to L1 in such a case.

If 64-bit, you may find little or no difference since Windows' own file cache cannot be fully disabled, though it can be bypassed for specific files by applications like benchmarking tools (see MS KB 99794: FILE_FLAG_WRITE_THROUGH and FILE_FLAG_NO_BUFFERING). Only if you run software that does bypass Windows' caching are you likely to see major performance improvements with PrimoCache.

With the configuration shown above, I'd suggest you'd be better off with one 8GB L1 for your most important disk (C: or D:) rather than 2 separate 4GB caches.

[Pliqui]

Hello InquiringMind

Thanks for the replay.

I'm using 32Gb ram and did what you mentioned the night i posted it. Now i'm a running a 10Gb cache job on my C and D drives with 5 minutes write deferred.

Sadly, i have a 32GB SSD laying around (well, it's a http://www.sandisk.com/products/ssd/sata/readycache/) But i ran out of SATA ports on my mobo.

Will keep that in mind when i upgrade my mobo to also to throw the L2 Cache.

Thanks again

[InquiringMind]

5 minutes write defer seems rather long - your mileage may vary but I found little benefit from exceeding the default 10 seconds and that is with regular backups to guard against data loss. If you want to keep a long write defer, I would suggest using versioning software (Aphar Backup is my preferred option - the webpage is Dutch but the program uses English - other options include AutoVer, Yadis Backup and the commercial FileHamster) to make automatic backups of your most important files (documents, savegames, etc). Aphar is my favourite since it is free and light (all the others require .NET Framework) but doesn't work fully with NTFS junctions (it will update files in a junction on startup, but won't detect further updates) though this can be worked around by "reversing" the junction.

Otherwise your configuration change seems a sensible one. If you're short of SATA ports, an interface card (or even a RocketRAID) may work out a cheaper option than a new motherboard.

[Pliqui]

Hello InquiringMing,

The 5 minutes write defer was just testing, agree with you no real benefit, now it's running at 20 sec.

Thanks for the tips on backup softwares, going to check those softwares. I do a monthly backup, weekly differentials and daily incrementals. Using comodo backup, so far so good.

Really appreciate your input and your time.

Thanks

My Final specs for primocache are the following

10Gb cache with 20 sec write defer Task with Block Size 8k and with 2 disks C and D

[Davey126]

One possible benefit to longer deferred writes is a significant reduction in the amount of data written to physical media. I typically see a 45-55% reduction with a 60s delay depending on work type. Three primary reasons: 1) temporary spool files (written to cache then immediately deleted); 2) redundant data (Primocache automatically discards); and 3) in-cache trims. I have all the appropriate power protections in place. That said, I have yet to suffer a catastrophic loss even on systems without a battery or UPS. Obviously there is some risk involved if you don't have the proper power protections but an unexpected crash does automatically equate to a corrupted system.

Given the resilience of today's SSDs does one really need to worry about managing/reducing writes in a typical consumer system? Probably not - but I love seeing Primocache metrics that show dozens of GBs of potential writes that never made it to the disk! Obviously reducing write activity does have a performance benefit on traditional media (HDD).