Weird SSD Benchmark

[idata]

Evertime I benchmark my intel 160gb G2 ssd drive I get a weird benchmark like this in the image. Is this normal?

[idata]

Your AS SSD result isn't very bad, just a bit off in 4K-64Thrd. Do you have the proper alignment?

To check that your partition is aligned correctly, go to your start menu and in the run box type msinfo32 and run it. Click on components => storage => disks and check what the partition starting offset is for your SSD. If it isn't divisible by 4096, then it's not aligned correctly (ie if you get any number that isn't a whole number, then it's not aligned).

[idata]

The result depicted in your benchmark is not weird--it is common for "physical disk" read tests on SSDs. A physical disk test does not use the file system at all, and can be run on formatted or unformatted disks with or without partitions or operating systems, or with a mix of them. In your case, I can tell from the shape of the test results that you have an operating system installed at the logical beginning of the SSD, that it is either Win Vista or Win7. and that you have probably run some file system read-write tests.

A physical disk test uses "absolute sector" addressing to talk to the disk. A typical data point will be generated by instructing the disk controller to read and return the data from sector number 1000 through sector number 1999. The test program will time the transfer, and divide the total number of bits by the total number of seconds for the operation to complete, in order to produce a single data point--then move on to, say, sector number 10,000 and repeat. It does not matter whether there is data in the test area, whether a sector may be part of a file, or whether a sector may be in a block of data on an SSD that has been marked "unavailable"--it all gets tested on an absolute sector read request. By contrast, on a file system test, all of this matters.

I am glad you posted on this point, because I am hoping that someone can answer my question--Why is the data transfer rate (read, not write) for an SSD slowed down so seriously in areas when there is actual data on the disk? And, the folllow-up question--Is any of the testing we have been doing on SSDs with test programs representative of the performance during actual usage?

I get physical disk test results very similar to yours using HD Tune and SiSoft Sandra. Here is a series of tests from Sandra over a period of one month.

Sandra provides a large database of test results from a wide variety of hardware and systems. Sandra also keeps your own prior results in the same database, and can display results from a current test, as well as your choice of four other results from the database. Sandra uses a small number of data points for the physical disk test compared to other test programs, and an unadjustable test block size of 10MB, so some of the finer detail is not there. Here is what the five traces represent.

The blue trace is for a Samsung disk and not relevant here--except for its odd shape. All of the other traces were from one Intel SSD as described in the window title.

The purple trace was taken when the disk had been in use a few days and the Win7 system was using the original driver for the drive controller. The remaining traces were taken using an RST driver. The large dip on the left is where the OS is installed. The first three or more % was the location of the hibernate file, which was never used, and that area tests high. The OS area tests low. There is a low dip in the center of the test--this is where Sandra places its test file for a file system test, and several of these tests had been run prior to generating the purple trace. There is a low dip at about the 20% mark which, I believe is the area used for AS SSD tests as well as for HD Tune's file system tests.

The green, gold, and red traces were taken using an RST driver. Note that the high results on these traces never get as high as the purple trace. The green trace was taken after the gold trace, and after quite a bit more testing with HD Tune.

Finally, the red trace was taken after I got curious about the SSD Toolbox, and ran its (extended, thorough, complete?) read-write diagnostic test on the disk. Note that essentially all of the flat-topped high-performance areas of the other traces are missing from the red trace, and it appears that read performance has been knocked down significantly in every area where data has been written. The Toolbox diagnostic test writes data into every free (by the file system) area of a disk and then reads the data back, and then apparently leaves the data there, but with no references to it in the master file table.

My point here is that the Intel SSDs test out at about the Intel spec for read performance only in places where there is no data on the disk. And in the data areas the performance is significantly lower. On jznomoney's test, you can see the read rate dip from 240 MB/sec to as low as 140 MB/sec in the data areas. On my less detailed tests the results are similar.

So, any comments on the significance of all this?

[idata]

I experience the same behaviour with a X25-V.

The drive had a perfect "line" in the graph and now I see serious drops of performance on some areas. This happened after I installed Windows 7 and then booted off it. I think the drops have nothing to do with the fact that Windows is running; it seems that the drops appear where actual data is "written" on the SSD.

That's weird.