Decreased speed with PrimoCache task

[Jaga]

Your software solution overhead CPU, and it sound almost the same as Windows pool management.

Not really. Windows storage spaces sucks compared to Drivepool. Tested them both in a production environment, won't ever go back to storage spaces.

You must have misunderstand me as I won't put any SSD into RAID... but let's start it once again.

Don't think I misunderstood - you mentioned "my plans are to RAID0 both of my current Toshiba P300 3TB drives, split into a couple of partitions to sort and separate bigger files like movies and other "big" media from all "smaller" like "my files" frequently used files, and buy a 6GB drive for backup only." Those are data/program drives yes? They'd benefit from Drivepool and it's solution. I wouldn't ever suggest putting your L2 into some sort of pooled solution.

Given the size of those two drives, and the fact they aren't the boot volume, I wouldn't put a L1 cache on them. Just a large L2, and keep your RAM for a L1 on the boot/application drive. So your assumption there was correct - don't try and use RAM for both boot and application/games/data volumes, just your boot volume. Trying to split it between them would mean a lot less cache space for each, and reduce hitrates.

Since your boot SSD will have a good sized L1 on it, it won't need a L2, meaning you can devote all your L2 space to your two spinner drives. Your choice to go either with RAID or a pooling solution - I gave up on RAID a while back after reading some white papers/articles on failure rates and realizing just how easily today's newer/larger drives can trash an array if just one bit doesn't pass parity checks. That's when I started looking at pooling solutions (~6-7 years back) and settled on Drivepool.

As for "..if L2 cache should only be configured to accelerate both HDD partitions, or maybe for accelerating another SSD too?" The answer is: never cache a SSD with another L2 SSD (unless as Support indicates it is much faster). The overhead will kill performance. Natively, there -is- no L1 cache "from a SSD", but with Primocache you can "tier" the two caches against the same volume(s) if you want.

Sounds like you've been lucky with drive failure rates in your career. I've seen quite a few go down, but then I work with clients (both end-users and companies) for all kinds of support, and used to work in a data center. Keeping my drives and data clean is top priority for me, and conventional backups are a must. The reason I use Drivepool and SnapRAID together is that I get the best of all features combined, and amazing flexibility. It's worth taking a look at since you haven't settled on a solution yet, and it has a free trial period I believe.

As an aside: Drivepool has very minimal overhead - it uses the partition's native format (i.e. NTFS) to do reading/writing. The software simply directs the OS to the proper volume depending on the file(s) needed.

[dondolarson]

Did you test them multiple times? Sometimes benchmark results vary much, especially when the speed is lower than 200MB/s. But if you use L2 to speed up HDDs, it will definitely speed up HDDs if read data is from L2 cache.

Yes, I did it many times. The results were always the same. As the L2 cache has been paused, everything was faster, not in benchmarks only. I've copied about 100GB movie files from and to this disk and usually, I've seen transfers as on the screenshots in my first post. I guess it's as you told before, L2 to a hard drive needs both writes and reads from it, not just reads as per default. I'll give it another shot later this year.

Not really. Windows storage spaces sucks compared to Drivepool. Tested them both in a production environment, won't ever go back to storage spaces.

I may give Drivepool a shot then, whenever RAID0 configuration fails for the first time for me So far I won't change my mind and stay with RAID0, especially when these 3TB drives are still covered by warranty.

I've just ordered ADATA XPG 8200SX Pro 1TB NVMe x4 SSD for system and games partition and bought a WD Red 4TB EFRX HDD for backups. I know I told it's gonna be a 6TB drive for backups but I won't exceed 4TB, I guess If so, then I'll sell it and buy a 6TB drive instead I've just sold the Intel 545s mentioned in my first post, so I can't test PrimoCache anymore since there's no SSD for L2 cache, and the trial is running out soon. I'll set everything up after reformatting my system later this year, and when I upgrade to 64GB of RAM.


I think that everything has been sorted out, except one question below.

If they're right and there is 0 L2 cache for a SSD, and with the RAM flush every restart, "prefetch last cache" and "start at windows boot" ticked, where's the L1 cache from an SSD stored then huh?

[Jaga]

You can email Support and as for a renewal on the trial once you get all the new hardware in place later on, they're usually pretty forgiving on that and willing to help.

As for your last question, "Prefetch last cache" looks up blocks to populate the L1 cache at boot from a block index Primocache keeps. The L1 cache isn't stored anywhere, just it's index is. Primo craws the index and reads all it's prior population info back into the cache when re-populating it (prefetching).

[dondolarson]

Does disabled Windows prefetching and indexing impact PrimoCache L1 (or L2 cache for that part) caching? These two functionalities are disabled on my PC and it's something I'm always doing right after re-installation of a Windows.

[Jaga]

I'd recommend disabling Windows prefetch, but if you use Windows Search then leave indexing on.

[dondolarson]

Guys... would you like to explain one more thing to me? Why is the cache hit rate decreasing instead of increasing? Weird. No changes made and I've seen 64, another time 40 and now just 14% and going lower... from 20 to 14 in 10 minutes Settings are 10GB L1 automatic r/w cache, block size 4KB, enabled defer-write 900s intelligent, prefetch last cache and a Windows Boot ticked. Accelerating my SSD with it.

[Support]

Cache Hit Rate = Cached Read / Total Read * 100%. So when the system or applications requests to read new data from source disks, the cache hit rate will decrease.

[Jaga]

Sounds like you are accessing a lot of different data on the target volumes, and have a cache size that can't comfortably accommodate holding the most used in it for a high hitrate. You'll need a larger cache task (L1, L2 or both) to handle all the data. I try to keep a "Data % Size" of about 20%, which means my cache task is 20% or more the size of the data on the target volumes. i.e. if I was caching 1TB of data, I'd want no less than 200GB of cache. That assumes that I frequently access a good portion of the data.

My hitrate changes daily depending on what I'm doing. I've seen it as low as 14%, and as high as 99%. Usually the more I add new files/programs to the drive the lower it gets, and the more I use those new files/programs the higher it rises. Different apps and data respond differently as well. As a rule of thumb, unless you use the same programs data over and over and rarely add new stuff, your hitrate will be lucky to be over 75% in my opinion.

[dondolarson]

That's what I'm gonna to test now. Thank you for your clear explanation.