Short Stroke Raid 0 F4 hybrid SSD software cache

Joined
Feb 23, 2005
Messages
650
I ordered 2 F4 320GB 7200RPM Samsung drives with relatively good reviews - faster than the Caviar Black 500GB for only $40. Access times and average transfer rates improve with short stroking to ~160G which is plenty for OS drive. Running it with a 60GB Vertex 2 and software hybrid cache to turn this setup into the seagate XT on steroids.

The short stroke RAID 0 F4 drives are the base drive at 80Gx2 = 160g of fastest portion of the drive. Software (Fancy Cache) runs L1 cache using 1GB RAM for frequently used data. it uses L2 Cache using 2 Gigs from the SSD for "secondary SSD Hybrid Cache".

This allows an already fast short stroke RAID0 array to become even faster applying 2 levels of cache on it beyond just the flimsy 16MB cache that comes on the drive already.

Any thoughts on this theory? Should have drives in to test in a few days. Running some tests on my WD 2TB green RAID 5 slowish arrays to see if benchies look good for now.
 
Ignore those too lazy or inept to try new things. Show us some benchmarks! One question though, why is it only using 2GB of the SSD?
 
well the main reason to do this is because lots of people have 60gb SSD's + hard drives for spill-over apps and storage. if we all had 256GB SSD boot drives it would be a moot point. it's actually rather easy to setup with a program called Fancy Cache that is in beta right now. Most people also use less RAM than they think they do or have 4+ gigs on a 32-bit OS so there is "free RAM" spare to use as cache or for RAM disks.

i've started testing it on my WD Green RAID 5 array short stroked to 80GB already since my F4 320's aren't in yet. On HD Tune I'm getting ~190MD/s average transfer rates and access time is 11ms which I think is ok considering these are about the slowest drives in existence. Crystal Mark goes nuts right now since too much of the data is getting hits in the RAM cache - over 5000MB/sec so I have to retest some things or disable the RAM cache portion.

the reason I started off with only 2GB of the SSD to cache with is because I only had 10GB left on my 60GB drive :) Hence the whole purpose of projects like this - people run out of SSD space and need to spill over to HDD and still want better performance. The Seagate XT uses only 4GB and gets a pretty huge boost, but I justed added another sandforce 60GB drive so I'll test more with 10GB of SSD cache and get some screenshots up when they make more sense.

btw if this sounds like a waste, apparently Intel doesn't think so since there are rumors they are imbedding SSD caching technology in their next set of RST drivers right?
 
apparently Intel doesn't think so since there are rumors they are imbedding SSD caching technology in their next set of RST drivers right?
First time I've heard of it. Gotta link?

You're short-stroking which will lose a lotta space and introducing RAID0 into the mix for increased fault possibilities.

Too much hassle, too much waste, too little benefit, but it'll be a nice project. :)

I'm looking forward to the results.
 
or have 4+ gigs on a 32-bit OS so there is "free RAM" spare to use as cache or for RAM disks.

usually this "extra" RAM in unaddressable by any program that operates at the OS level or under. ive never used "fancy cache"; does it have hardware level access so that it can address RAM over 4gb on a 32 bit system?

im interested to see the performance numbers you get on this. its certainly a worthwhile project to tinker with, though i would have some strong concerns of reliability from a "franken-RAID" array like this. it would be almost impossible to recover anything from a drive failure, and with the amount of RAM/caching you are doing, a power failure or blue screen could be a disaster. i dont know that your gains are going to be worth going this route over just a simple SSD (or two) for an OS drive and a HDD data drive like is so popular.
 
"The one feature that looks really interesting is RST SSD Caching, where RST stands for Rapid Storage Technology. If this is what we think it is, it'll allow you to use an SSD as a cache drive while using a hard drive as the main storage drive. We've seen solutions like this in the market via various hardware solutions already and it's not too farfetched that Intel would integrate such a feature in its chipsets. With RST driver version 10.5 Intel will also support RAID with hard drives larger than 2.2TB. What's missing is native USB 3.0 support, so it seems like Z68 based motherboards will still have to rely on third party USB 3.0 host controllers."

http://www.semiaccurate.com/2010/11/16/intel-z68-chipset-rumours-start/

Fancy Cache does have hardware level access so you can access RAM over 4GB.

I realize this isn't the most practical thing to do but that's what we do for "fun" :) I have several SSD's, with almost every popular controller just for testing/experimentation so no actual true lack of fast storage. The concept of using SSD as cache space for slower HDD's has many implications beyond getting by with the simplest solution (get a larger SSD when you run out of space).
 
Dude, just get an SSD. The hell are you thinking? 320GB drives?

+1

Yeah, for so little storage, go with a SSD.

Also, you can't short-stroke a HDD by taking only 50%, you need to take 70-80% of the drive space to even make it worth it.

Seriously, a mid-range SSD can be had for sub-$100 now and will blow those things away no matter how many you RAID together.

But since you already got them, good luck. :p
 
Ok some possibly interesting updates. I decided to throw in a spare 60GB Sandfore SSD and Seagate 1.5 TB freshly formatted HDD for testing since it's more practical to test just a typical SSD Bootdrive + large storage drive combination. More people were interested in the idea of a software cache using SSD space anyways such as the rumors for the next Intel RST drivers and chipset. My initial idea of testing cheap/fast RAID 0 HDD's was more out of curiosity than a real need for fast storage space since I have enough spare SSD's purchased just for testing. I wanted to see if two fast HDD's could approach SSD-like speeds using "free RAM" cache above the 32-bit OS limit in addition to a small amount of SSD space, but apparently I'm the only one curious about this :) Whether or not a 2 level software cache using free RAM and SSD cache can significantly speed up your storage drive is the more practical question.

PCMark Vantage productivity HDD suite:

this is the most interesting one since it shows that the SSD Cache through "Fancy Cache" does something. This is a long, 20 minute suite of multiple hard drive tests that are supposed to mimic "real use" scenarios such as importing hundreds of photos into Windows Gallery, media streaming, TV recording, application and game load times, etc.

The first number is just the bare drive, then with SSD cache first turned on for initial run, then for a second run where there is actually some caching being done between the first and second run. The minimal 128MB L1 RAM cache is used and 10GB SSD space is used for L2 (The program won't allow 0 L1 RAM cache).

4527 --> 4663 --> 18926 (pure Sandforce SSD score is 40k)

so you can see that the bare drive with no cache added is horrible. when you turn on the SSD cache and run it again none of the data has been cached yet so it's essentially the same horribleness. however if you do a SECOND run with SSD cache on, you get a gigantic increase to a performance number close to halfway between the HDD and the SSD drive. you might be able to extrapolate from this and assume that if a user were on his computer for a few hours at a time and switching between the same programs, there could be a large impact on performance. if he is only using each program once and never going back to it, the gains would be small.

HD Tune Pro

Avr Transfer Rate 81 MB/s Access time 16.3
Cache has no impact on HD Tune Pro and the numbers stay the same no matter how much cache is thrown at it. The cache log does not even record any data being written during HD Tune tests so they somehow do low level read/write that can't be influenced with software Cache.

ChrystalMark

Chrsytal Mark fully caches all it's data so this info is posted only for amusement. The software cache program I had required minimum 128MB of RAM cached be used in addition to the L2 SSD cache. The RAM cache blew up these numbers into huge proportions and doesn't really mean anything, but here it is.

Over 7000 MB/s random writes and 1000+MB/s 4k Random Reads!
Sequential Read : 5743.238 MB/s
Sequential Write : 4129.065 MB/s
Random Read 512KB : 5534.539 MB/s
Random Write 512KB : 7184.246 MB/s
Random Read 4KB (QD=1) : 492.776 MB/s [120306.6 IOPS]
Random Write 4KB (QD=1) : 612.949 MB/s [149645.8 IOPS]
Random Read 4KB (QD=32) : 1125.094 MB/s [274681.3 IOPS]
Random Write 4KB (QD=32) : 977.015 MB/s [238528.9 IOPS]


kcc8j8.jpg
 
I think it's great how you have made the most of your storage drives. I had in mind something similar. Hope my set up works with my Z68 MSI GD65 G3 MB.
 
Back
Top