Messing around with RAIDed SSDs...

I'll post a benchmark when I get home. I'm pretty sure my low numbers are due to using it in a Dell Vostro 3300 laptop (2 to 3 years old). CPU is Intel Core i5 520m and Chipset Intel HM57. Samsung Magician software indicates it doesn't support SATA 3 (and by looking at the intel website, it seems the chipset only supports up to SATA 2; even so, Dell may have chosen to use a lower speed or something from what I read).
I did improve my numbers a bit by doing some basic tweaking in Windows for SSDs.

 

You do know that his cheap pciex1 "raid" card is probably not a hardware raid card persay? ( aka fakeraid or whatever other terms are thrown around for it )

Those in general have terrible performance.

sonicdude10: could you snap a pic or tell us the chipset on your raid card?

 

Yes, I am well aware - I work with server hardware and expensive raid controllers daily.

I was supposed to say the pci-e card wasnt proper raid, not onboard. But onboard and these cheap controllers are the same thing.

 

I can't right now as I don't feel like opening up the case until I get my new motherboard but as was said earlier, I don't think it is "true" hardware RAID. After all it was only $18 on FleaBay...

I guess it is just a small standalone setup with dedicated BIOS to communicate with PC for the data striping...

 

Here's my benchmark:

 

2x Samsung 840 pro 120Gb Raid0 @ GA-Z77xUP7

 

Pretty much the same as mine, except for the 512K writes.

Speeds on mine were higher initially, but my chipset doesnt support TRIM on raid sets. I later hacked the bios to add support, but I need to erase my drives and start over for max performance. Or just wait until trim eventually fixes it.

 

I find it funny how the 4K writes are much higher than the reads. I thought most drives have higher read speeds than write. Maybe it has something to do with searching for the files needed.

 

Hate to resurrect this but thought I'd share something interesting about that PCI E X1 2 port SATA expander card I have. I was using it to copy some data from 1 HDD to another. Interesting thing is that one drive was hooked to the motherboard SATA port and the other drive to one of the ports on the card. My average transfer speeds hovered around 45 MB per second. More transfers done with direct motherboard SATA port to port showed an average speed of around 20 to 22. Interesting.

On a side note, I couldn't get my system partition moved to the X2 SSD drive so I decided the next best thing was readyboost. Split the drive into 4 partitions. First 3 are just under 30GB and last just over 23GB. Filled all 4 partitions with readyboost cache files to the brim. Only 21MB is left free because readyboost always leaves around 5MB of a drive untouched. System has sped up DRAMATICALLY since my most common applications are now being cached to the 2 SSD drives instead of the HDD.

Readyboost on multiple RAID 0 SSD partitions: the poor man's way to get SSD caching...

 

I don't have an exact answer for your speed differences, but i have experienced this before on on fiber channel RAID storage. I have always assumed it was because the controller was reading and writing at the same time and you end up saturating the controller.

When you have two controllers it is a one way transaction.

 

ReadyBoost on SSDs is the rich man's way to get SSD caching. It will kill those drives. Better to add more RAM into your PC instead.

 

The drives were cheap. Adding more RAM won't speed up the hard drive. The readyboost caching works by taking the most commonly used Windows files and having them stored in the cache files so they're ready to be accessed right away. More RAM won't help unless I made a RAM disk and put a readyboost cache file on that. Not going to do that...

 

Actually you should add a ram disk, but considering the drives are cheap there isn't a real reason to change. The will die a much quicker death this way though.

 

Windows Prefetch, SuperFetch and ReadyBoot does exactly what you are describing, but with RAM.

Unused ram is wasted ram, so it fills it with things it thinks you will use. If you need ram for something else, it can immediately drop that contents and use it for something else. This is why peoples obsession with "unused ram" is stupid.

Also, readyboost on SSD is not sensible at all.

 

I know it isn't sensible but the drives were dirt cheap so them getting killed faster won't matter to me. If I remember the tech specs from years ago it was stated these models could handle upwards of 20 or more gigs being written to them constantly 24/7 for like 6 or 7 years. I'm sure I won't be using them even in the next 3 years so all is good.

 

To estimate the lifetime of a ssd, search its maximum "wear-levelling count", i.e. the number of times it can be written over before it's dead. For TLC drives it's around 1000, for MLC it's typically 3000, for SLC it's even more.

Now do this:

T = (C*W/(A*D))/365.25

Where:
C = Capacity of drive in GB
W = Maximum wear levelling count
A = Write amplification (~3 to 10)
D = Average amount of data written per day on the drive
T = Lifetime of the drive in years

This is a gross estimate, because A varies a lot, it's around ~3 for normal usage; ~10 for random writes of small files.

Take my own drive as an example, it's a samsung 840pro. It's a C=256GB drive with W=3000, writing an average D=20GB per days and amplification ~3, you get ~35 years of lifespan, which ain't bad.

Cheers!

FG