SSD performance difference in relation to disk occupation

Silencing hard drives, optical drives and other storage devices

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
proc
Posts: 47
Joined: Thu Sep 07, 2006 4:39 am
Location: Italy

SSD performance difference in relation to disk occupation

Post by proc » Fri Sep 19, 2008 3:48 am

Hi,

I think this is an import thing to share: a "factory clean" SSD is much faster than a disk where data has been written to!!

in this Italian review http://www.hwupgrade.it/articoli/storag ... se_10.html you can see the graphs of the different reading and writing speed when the SSD benchmarked is empty and when is full of data (in this case written by I/Ometer test).

In particular it's interesting the case of the Intel X25-M:

this is the result of the first iteration of I/Ometer on the disks: http://www.hwupgrade.it/articoli/storag ... _prima.png

and this is after some runs of I/Ometer:
http://www.hwupgrade.it/articoli/storag ... r_dopo.png

The difference between the two graphs is bigger for the Intel drive since it was just sent by Intel (factory clean) while the other two disks were already tested and thus not "clean" anymore.

The journalist contacted Intel to ask if they had a tool to reset the original state of the disk, and he was able to reset the disk with it.
http://www.hwupgrade.it/articoli/storag ... se_11.html


The article correctly states that the reset tool is useless for everyday use, but the performance difference between testing conditions and real world usage is something to keep in mind![/b]

Aris
Posts: 2299
Joined: Mon Dec 15, 2003 10:29 am
Location: Bellevue, Nebraska
Contact:

Post by Aris » Fri Sep 19, 2008 9:58 am

i cant remember where i saw it, but i saw an english review that noticed the exact same thing on the new intel MLC drive.

They did all the regular testing, and then at the very end, they re-ran one of the tests but with the drive at 80% capacity filled. It was a compilation test, that tested MB/s throughput for various applications. Half were the same within the margin of error, but the other half took like a 30-50% performance hit. That said though, even at a reduced speed, it was still like 4 times as fast as a velociraptor. So you have to keep everything in context.

Moogles
Posts: 315
Joined: Thu Mar 22, 2007 10:28 am

Post by Moogles » Fri Sep 19, 2008 12:03 pm

I benched my 16GB Mtron mobi (SLC nand) when I first got it and got a perfectly even sequential read speed line in HDTach and HDtune (101MB/s). Now that it's been in use for a while, even though it's not noticably slower in any regard, I don't get a perfectly straight line in benches anymore, and I've taken a small speed hit.

Ehh, it's still so much better than any conventional harddrive, I'm not really bothered by it.

proc
Posts: 47
Joined: Thu Sep 07, 2006 4:39 am
Location: Italy

Post by proc » Sun Sep 21, 2008 4:30 am

I agree with both of you regarding performance, especially sequential writes are not a problem, but random ones could be (see the Core's freeze).

I was just pointing it out because this reduction in performance after some use is something to consider when we see comparison of several SSDs with some that could be brand new and others already used.

Moreover I think good reviews should generally test hardware equipment for the medium and worst case scenarios to let the reader know what to expect.

Aris
Posts: 2299
Joined: Mon Dec 15, 2003 10:29 am
Location: Bellevue, Nebraska
Contact:

Post by Aris » Sun Sep 21, 2008 8:00 pm

I dont think this is a generally well known phenominon. SSD's are still fairly new. I would highly doubt most reviewers of SSDs even know about this issue, let alone to test for it.

Turas
Posts: 43
Joined: Thu Aug 02, 2007 9:29 am
Location: Alexandria, VA

Post by Turas » Mon Sep 22, 2008 6:10 am

Interesting find, I will have to keep a eye out for more about this. I am not totally shocked by the MLC one though but am surprised that the posters above SLC drive expereinces this also. Since the R/W cycles on the MLC flash is a lot less I am guessing there is more error correction/tracking of which flash has been used then there is on the SLC ones.

Post Reply