[SOLVED] Poor SSD Performance, Low Buffered Disk Reads

Question

Hello. Today, I ran hdparm -t and noticed unusually low performance:
/dev/sdb:
Timing buffered disk reads: 342 MB in 3.00 seconds = 113.85 MB/sec
This is on a Samsung 840 EVO with an FX-8320 4.5ghz. Testing /dev/sda, my mechanical hard drive, gives me nearly identical results, which makes pretty much no sense. I ran this about a week ago, and got numbers around 520 MB.
I am using the BFQ scheduler and the Linux-ck kernel. I tried other IO schedulers, but only gotten worsened performance.
The only thing I can think of that I've changed recently was from the performance to ondemand CPU governor. Could that have caused this?
Any help would be appreciated. I am using UEFI/GPT. uname -a:
Linux msw1 3.13.11-2-ck #1 SMP PREEMPT Fri Apr 25 20:19:47 PDT 2014 x86_64 GNU/Linux
EDIT:
Now solved! It turns out the 840 EVO with its TLC NAND is an odd SSD, and that the alignment needs to be a multiple of 1.5.
Last edited by msw1 (2014-05-01 21:28:31)

Answer 1

supermariolinux wrote:
I have the same SSD(840 EVO) as you do and i am getting similar results with hdparm -t, although i thought this had something to do with the fact that my SSD is only connected to SATA 2.
Partitions are aligned and manually trimming the partition doesn't seem to change anything.
I noticed there is  new firmware update available on Samsung's website(You can boot from a CD with the ISO without having to use Windows), unfortunately i cannot get it to work as the program just crashes.
Using the standard 3.14.1-1 x86-64 kernel with default cfq scheduler
EDIT: It appears i already have the latest firmware installed, still i recommend checking with
hdparm -i /dev/sdX | grep FwRev
and seeing if new firmware is availbe here: http://www.samsung.com/global/business/ … oads.html#
hparm -t results:
/dev/sdc:
Timing buffered disk reads: 368 MB in 3.01 seconds = 122.36 MB/sec
I do have the latest firmware. As for you, it could be SATA2. I, however, have this connected to a SATA3 controller. As I said, this used to get 500+ easily. Now, it's more like 80.
gedgon wrote:I had exactly the same issue with my EVO when it was not aligned to EBS. It's 1536 KiB or 2048 KiB (depending of source), so it's safe to set the sector alignment value to 12288 (6144 KiB), multiple of 1536 KiB and 2048 KiB.
I wish this solved it. However, I still get the same result:
sudo blockdev --getalignoff /dev/sdb
0