my OCZ Core SSD analysis

hmm interesting info, there is definitely something wrong with the write side of that drive.
 
Great analysis, better than most of the previews/reviews of the OCZ Core series out there. Convinced me to wait a few more months to purchase any kind of SSD.
 
what services write a lot to the disk since a list would be a good way to tweak the os.

Seems its worth it due to a snappier desktop tho.
 
I think the main problem is that you can't use harddisk benchmark tools on flash media.

You would have to get detailed technical data for the SSD to know what you should test and what you should not test.

Flash memory can only be erased in certain block sizes. If you write less than that, the controller has to read out the whole block, modify a part of it, flash the block and write it back. This means that you need to know the erase size unit. If you use a filesystem, clusters must be at least the size of the erase unit and have to be properly aligned with those. Some controllers are modified for a certain offset, because FAT filesystems usually have cluster 2 on and odd sector. If you use benchmarking tools that start at sector 0, you might force the flash controller into unaligned writes.
 
main problem? what problem?

the drive is well understood...

did you read my article?
 
8steve8 said:
main problem? what problem?

the drive is well understood...

did you read my article?
Click to expand...

I was refering to the irregular write results, without mentioning that, sorry.

Harddisk benchmark tools assume that writes can be done in 512-byte increments at random locations without affecting the results. This is not true for most flash media.
 
hjreggel said:
I was refering to the irregular write results, without mentioning that, sorry.

Harddisk benchmark tools assume that writes can be done in 512-byte increments at random locations without affecting the results. This is not true for most flash media.
Click to expand...

What's your point? Are you saying the results of the benchmarking tools are somehow invalid because the results clearly illuminate the problem with NAND flash's random write performance?
 
Hey 8steve8, did you happen to receive any messages from the OCZ staff regarding your post at the OCZ forums? Generally the OCZ staff responds quickly to posts made on their forums.
 
known12345, that isn't my post, but it does link to my article.


by the way i re-wrote the article a bit, adding analysis explaining the writes.
 
Kaldskryke said:
What's your point? Are you saying the results of the benchmarking tools are somehow invalid because the results clearly illuminate the problem with NAND flash's random write performance?
Click to expand...

My point is that there is no NAND flash random write performance problem, as long as you respect the internal structure of the flash memory, just as you accept that harddisks come with a sector size of 512 bytes.

It would be the same as if you would compare a RAM disk and a harddisk by reading and writing data the size of a cache line of the main memory at random "even" addresses. It's just that you can't write to arbitrary positions on a harddisk, so this comparison is impossible.

An SSD tries to be smart by *pretending* to have 512 bytes per sector, while the internal structure is based on much larger units. It just won't tell you how much it hurts (performance-wise) if you don't write a whole recording unit at a time. If you set up a filesystem with 32kB cluster size aligned to the recording units, the SSD will most likely have a much better real-world write performance than reported by the benchmark tools.
 
hjreggel I agree with you for the most part, but i discuss this in my article.
(it's been heavily updated in the past 24 hours)

IT's not that the tools have a problem, in fact all the results make sense, i think i explain it all but yes it has to do with what some call erase blocks, which effectively incur random writes with a time-cost that is not constant, depends how much of the start and end block/cluster is data relevant to the write....
 
I noticed you used the ICH9R controller. Is there any chance you could use a 3rd party controller to confirm your results such as an Areca?
 
If you set up a filesystem with 32kB cluster size aligned to the recording units, the SSD will most likely have a much better real-world write performance than reported by the benchmark tools.
Click to expand...

Is this something that is easily doable with a normal install of XP or Vista?

And, although the write access time is horrible, how badly does it effect real world usage?

I suppose I'm gonna hafta go to the OCZ forums for some indepth real world observations.

I sure wish these things had been hunky-dory. My CC was ready for a work-out! :D
 
setting ntfs cluster size to 32kB wont do anything significant regarding this ssd's performance.
 
the ssd's clust size is much larger than 32kB, like 4-16MiB.

if you would set the ntfs cluster size to 4MiB, then you might see something, but would you really want to do that, even if ntfs would allow it?
 
Excellent read. Thanks for taking the time to write that up.

So with write performance being quite sub-par (compared to mechanical drives at least), these still make for good OS drives (assuming no page file, etc), right?
 
yeah, After installing vista, the boot drive is mainly used to read from..(not write to)... there are writes to be sure, but not to the extent where you should demand high random write performance... of couse we still want it, but this ssd should be a lot faster than any rotating disc hard drive for a normal boot drive without page file/system protection/drive-write-caching,
obviously disable the automated defrager...
 
you mean the updated ones with good write speeds and similar prices to the OCZ?

I was under the impression that all these new affordable SSDs that have arrived on the market are re-badged Samsung's, but I can not garentee that.


this includes the RiDATA one as well...
 
You must log in or register to reply here.
Back
Top