Accelerating PrimoRAMDisk with PrimoCache by 100%!

[Axel Mertes]

You need to open a DOS box in windows, such as entering "CMD" in either the run/start or windows file explorer.

Then enter:
fsutil fsinfo ntfsinfo <DRIVE>:

example:

C:\Windows\system32>fsutil fsinfo ntfsinfo c:
NTFS Volume Serial Number : 0xf4ca5d7cca5d3c54
Version : 3.1
Number Sectors : 0x00000000378fd7ff
Total Clusters : 0x0000000006f1faff
Free Clusters : 0x00000000000e8821
Total Reserved : 0x0000000000000910
Bytes Per Sector : 512
Bytes Per Physical Sector : 512
Bytes Per Cluster : 4096
Bytes Per FileRecord Segment : 1024
Clusters Per FileRecord Segment : 0
Mft Valid Data Length : 0x00000000196c0000
Mft Start Lcn : 0x00000000000c0000
Mft2 Start Lcn : 0x000000000097ffff
Mft Zone Start : 0x000000000051f920
Mft Zone End : 0x000000000051f9a0
RM Identifier: 0652C3D3-7AA9-11DA-ACAC-C80AA9F2FF32

The cluster size is what are looking for.

[InquiringMind]

I did some testing myself using Anvil (since AS SSD requires .NET Framework) and came up with different results. Firstly, my system is based on a Gigabyte X58 motherboard with a Xeon 5650 processor (6-core plus Hyperthreading) clocked at 3.4GHz and 18GB DDR3 RAM clocked at 1.7GHz. Principal storage is a RAID-0 array of 6 Samsung SSD's. I'm running WinXP SP3 (and therefore using "invisible memory" above the 3.25GB manageable limit for both PrimoCache and PrimoRamdisk).

First for a baseline, here are my benchmarks for the SSD array with and without PrimoCache (read/write, 64KB block size, 10-second write defer).

[InquiringMind]

First of all, I disabled my existing (6908MB) cache and used 2GB to create a second ramdisk (my existing one has the Windows pagefile and temp files on it making its deletion tricker) in both SCSI and Direct IO versions.

[InquiringMind]

In both cases, ramdisk performance was not far off that provided by PrimoCache, with DirectIO being slightly faster for sequential access and SCSI doing better for queued transactions (at least with queue depth 4).

Next I set up PrimoCache on the SCSI ramdisk - only 2812MB RAM was available for the cache due to the second ramdisk, but with Anvil using 1GB data for testing, that should be enough. I used default settings as much as possible but ran tests twice - the first with a 4KB block size and the second with 64KB blocks (other settings: read/write, 10-second write defer).

[InquiringMind]

In this case, PrimoCache provides a small (barely noticeable) boost to sequential reads and a small reduction on queued reads. The standout result was that with 4KB blocks, write performance dropped by two-thirds (and this was the best - other runs gave 2400-2800MB/s sequential writes, making this a highly variable result in contrast to all the other tests).

Honestly I would have expected the both apps to be similar in speed, with the RAM Disk potentially being faster as to not needing to worry about looking up cached blocks etc.

Your system is running the same CPU as mine but at a lower speed, memory ditto so I would expect your results to be about 2/3 of mine (as PrimoCache/Ramdisk performance scales linearly with memory/CPU) - so around 6,000MB/s sequential. So the question should be, why is your ramdisk so slow?

However, it turns out the opposite is true. In my opinion PrimoCache completely replaces PrimoRAMDisk.

Not really - PrimoCache can help with repeated reads/writes so it's actual performance will depend on the application used, with disk benchmarks showing it very much at its best. Primo Ramdisk on the other hand should provide more consistent performance benefits in all situations but requires more setup work to identify and relocate files.

[Axel Mertes]

While I basically think that the RAM Disk should be faster per se than PrimoCache, its completely repeatable among different benchmark tools. I don't know if the measurement fails "by design" or what causes this behaviour. However, I have made a bunch of tests showing it and its surely worth investigating, as it must have a reason why this happens. Maybe code can be optimized, maybe its some kind of rare circumstances. I don't know, I was as much surprised as anyone else. It was so absurd in first place, I just gave it a shot and was blown away with the results I saw.

Looking forward to PrimoCache3 with improved caching options will allow me to run at least about as fast for my use as a RAM Disk would be, but way more flexible.

I now run our company server using PrimoCacheServer and drives have never been so reactive before. If we get to the point that we can deal with reliable failure safe write caching (some kind of journaling plus potentially battery buffered flash memory, with recovery option) then we should almost never feel the slow speed of HDDs anymore.

A true RAM Disk is of rare use for me, usually too small, too complicated. A cached drive is almost unlimited in size and way easier to maintain. But there are scenarios for which a RAM Disk is perfect (such as browser cache or the like).