Level-2 cache alogrithm in version 0.9.8

[Support]

There're three major changes in the level-2 cache alrogrithm in version 0.9.8.
1. PrimoCache stores disk (being cached) data to level-2 storages when computer is idle. In previous versions, it did when level-1 cache is full.
2. PrimoCache keeps frequency of use data and stores frequently used data to level-2 storages. Currently data block with frequency larger than 1 will be stored. That means a data block has to be requested at least twice before it can be stored to level-2 storages. There's one exception. Boot data will be always cached reglardless of frequency.
3. Only read-data can be stored to level-2 storages regardless of cache strategy.

Here is an example to show how level-2 cache works in version 0.9.8.
1) Setup Cache
Create a cache task on the boot Windows partition with the following settings:
Level-1 Size: 0MB
Level-2 Size: 10240MB
Block Size: 4KB
Strategy: Read-data & Write-data

pcboot1.JPG (48.6 KiB) Viewed 7004 times

2) Restart Computer
During the boot phase, PrimoCache traces the data usage. After boot is ready and system becomes idle, PrimoCache will store frequently used boot data to the level2-cache. You may check the “L2Storage Write” value listed in the statistics to know how many bytes PrimoCache has stored to the level-2 storage. Here we leave the computer Idle for certain seconds after we log on to Windows. And we'll see that almost all boot data will be stored to the level-2 storage.

pcboot2.JPG (109 KiB) Viewed 7004 times

3) Restart Computer Again
Restart the computer again and we see that PrimoCache has provided almost all needed boot data from its level-2 cache. The cache hit rate is near to 100%.

pcboot3.JPG (107.46 KiB) Viewed 7004 times

[Bjameson]

I tried this with Windows 8.1 x64 and YES! It works! I had tried before but I didn't see it cache anything. That was because I had no idea that Primo delays L2 writes. Now that you explained how it works, I have a better understanding of your algorithms. I now get cache hit rates of over 97%.

My boot volume is an SSD and I created a 4 GB L2 cache on another SSD. This makes the whole thing even faster. Without Primo, Windows must read & write at the same time to the same SSD. With Primo, Windows reads from one SSD and writes to the other. Yes I could delay those writes through a RAM cache but experiments like this are fun.

I have tried all other options and so far I think Primocache 0.9.8 works great & is safe to use.

Maybe the next step would be to make Primo ready for the market. How about colorful animations so that users can visualize what Primo is doing? Look at Supercache, O&O Clevercache to get an idea of "wrapping" your superior product into a nice, attractive package that users want to buy.

Please keep those technical explanations coming. They're extremely useful. Thank you!

[minhgi]

If the dirty write delay is set to infinite, it looks like the data does not get cache that much. But if the delay is set to 10 seconds or least, than the data get cache more often.

Also does the write cache data is store still in the L1 or L2?

[Jaga]

Achieved 95.68% cache hit rate using this method, which includes a 20gb RAMdrive mounted prior to Windows startup (a boot drive for temp files).

[piquadrat]

There're three major changes in the level-2 cache alrogrithm in version 0.9.8.

2. PrimoCache keeps frequency of use data and stores frequently used data to level-2 storages. Currently data block with frequency larger than 1 will be stored. That means a data block has to be requested at least twice before it can be stored to level-2 storages. There's one exception. Boot data will be always cached reglardless of frequency.

How the caching algorithm knows which data is boot and which not if it works on block level?

I have done some tests.
Win 7 64-bit Ultimate, 16GB RAM, Crucial SSD 256GB used as the system drive and Cache Partition (28GB with 26GB of cache volume), RAID-5 (D: drive) array set up as a target which is supposed to be cached. Settings are exactly the same as in your example. Only L2 enabled, L1 set as 0MB.
Just after defining the Cache Task for D:[Programs] drive:



After running 1 particular application from D: drive 4 times in a row:



After restarting the system and wait for 10 minutes on Desktop:



After running 1 (still the same) particular application from D: drive 4 times in a row:



After restarting the system and wait for 10 minutes on Desktop:



And finally running 1 (still the same) particular application from D: drive 4 times in a row:



Each application run reads around 1GB from the D [Programs] drive. Usually the first run is read from the drive, the rest from windows disk cache I suppose. No other application was being run from D: drive in this test.
As you can see virtually no data was being cached. FreeCache (L1/L2) value still reads 0/24GB.
Comparing L2 Storage Read before and after application run indicates no Cache Read at all.
No speed improvement.
Maybe L2 cache works only for boot/system drive?

[Nina]

@Romex .. Fantastic work! L2 Cache works great! - I really like the new algorithm.. No cache pollution at all! Superb!.

[minhgi]

I confirm that it is work. One question, does this pertain only to small frequently access data or large data does get cache as well?

[minhgi]

Any chances of beta tester getting a free license for Primocache when it become final? I have been using since its first initial release and giving feedbacks since then. Hehe...just asking. Its worth a try.

[InquiringMind]

...data to level-2 storages...

One small thing - the UI and instructions should refer to "level 2 storage" not "level 2 storages". "Storage" is a noun where the plural is the same as the singular.

It's a minor issue, but the only English mistake in the UI I've come across.