Fedora 15 Sandy Bridge SSD speed regression 2.6.38 / 2.6.40

[pepousak]

Fedora 15 / Lenovo T420 / Sandy Bridge / SSD Crucial 256GB nearly 500 MB/Sec working perfectly with kernel 2.6.38, after upgrading to 2.6.40 (due to external monitor problem), speed decreased to 300 MB/Sec. Thanks for any ideas

[pepousak]

update to 2.6.40.3 doesn't help

[oldcpu]

Does this boot code help ?

i915.semaphores=1

[pepousak]

Originally Posted by oldcpu

Does this boot code help ?

i915.semaphores=1

Unfortunately not. I did the same test (F15, 2.6.38/2.6.40.3) at a Non-SandyBridge Netbook HP Mini with the same result. So it's probably not related to SandyBridge.

[stevea]

Can you post more details about the T420 and the crucial drive model ? How are you measuring speed ?

So far I have found no crucial SSD /SATA-3 that claims more than 415MBPS in sustained sequential reads.
FWIW the Micron/Crucial M4 (C400) series (which I assume you have) are pretty sweet.
But the technical review here shows 404.8 MBPSon sequential reads - close to the mfgr's 415 MBPS claim.
http://www.storagereview.com/crucial...d_review_256gb

---

To get 500 MBPS or even 400, you'd need SATA-3, but a limit of ~300MBPS is about where SATA-2 stops. So this brings into question exactly how your interface is configured in the two kernels.

I *think* that if you do this command on both OSes you might see a difference.
sudo hdparm -I /dev/sda
Go ahead and post the results (I'd like to see 'em - curiousity).

My SATA-2 drive has the following lines of output ...

* Gen1 signaling speed (1.5Gb/s)
* Gen2 signaling speed (3.0Gb/s)

and I expect you will get a 6Gb/s line. Note that the '*' asterisk indicates the feature is enables, or available.

Have you tried a 3.1 kernel ?

---------- Post added at 04:21 AM ---------- Previous post was at 04:04 AM ----------

I just found that you should gt a message ....

* Gen3 signaling speed (6.0Gb/s)

But the F14 hdparm is too old to contain this. You need
hdparm-9.36-2.fc15.x86_64
not
hdparm-9.27-1.fc13.x86_64

[pepousak]

Originally Posted by stevea

Can you post more details about the T420 and the crucial drive model ? How are you measuring speed ?

So far I have found no crucial SSD /SATA-3 that claims more than 415MBPS in sustained sequential reads.
FWIW the Micron/Crucial M4 (C400) series (which I assume you have) are pretty sweet.
But the technical review here shows 404.8 MBPSon sequential reads - close to the mfgr's 415 MBPS claim.
http://www.storagereview.com/crucial...d_review_256gb

---

To get 500 MBPS or even 400, you'd need SATA-3, but a limit of ~300MBPS is about where SATA-2 stops. So this brings into question exactly how your interface is configured in the two kernels.

I *think* that if you do this command on both OSes you might see a difference.
sudo hdparm -I /dev/sda
Go ahead and post the results (I'd like to see 'em - curiousity).

My SATA-2 drive has the following lines of output ...

and I expect you will get a 6Gb/s line. Note that the '*' asterisk indicates the feature is enables, or available.

Have you tried a 3.1 kernel ?

---------- Post added at 04:21 AM ---------- Previous post was at 04:04 AM ----------

I just found that you should gt a message ....


But the F14 hdparm is too old to contain this. You need
hdparm-9.36-2.fc15.x86_64
not
hdparm-9.27-1.fc13.x86_64

Thanks for tips, I did some tests with hdparm and all 3 modes are indicated in both kernels:
* Gen1 signaling speed (1.5Gb/s)
* Gen2 signaling speed (3.0Gb/s)
* Gen3 signaling speed (6.0Gb/s)

I run palimpsest Read-Only Benchmark (gnome-disk-utility-3.0.0-2.fc15.x86_64) and it looks like 6.0Gb/s is working in both kernels. Average Read Rates:
2.6.38.8: 526 MB/s
2.6.40.3: 496 MB/s
which is nearly the same and I don't care about this tiny difference.

However, when I use dd or my real-world C modules for sequential read there is a noticeable difference:
2.6.38.8:
dd if=/dev/sda2 of=/dev/null bs=1G count=8
589934592 bytes (8.6 GB) copied, 15.4287 s, 557 MB/s
2.6.40.3:
dd if=/dev/sda2 of=/dev/null bs=1G count=8
8589934592 bytes (8.6 GB) copied, 27.9272 s, 308 MB/s

My SSD is Crucial M4 256GB and I'm very happy with this model.
Once more thanks for your effort.

[pepousak]

While exploring the issue I've written a C test sequentially reading 16GB from /dev/sda (I've got 8GB RAM). I've tried various block sizes and finally the difference came out. The read throughput is nearly the same in both kernels when the used block size is from 1K to 128K. Starting from 256K I can see a big drop in the kernel 2.6.40.3.

Kernel 2.6.40.3:
353 MB/sec 16G / 1K
354 MB/sec 16G / 2K
353 MB/sec 16G / 4K
354 MB/sec 16G / 8K
354 MB/sec 16G / 16K
353 MB/sec 16G / 32K
353 MB/sec 16G / 64K
353 MB/sec 16G / 128K
260 MB/sec 16G / 256K
260 MB/sec 16G / 512K
260 MB/sec 16G / 1M
260 MB/sec 16G / 2M
258 MB/sec 16G / 4M
259 MB/sec 16G / 8M
258 MB/sec 16G / 16M
258 MB/sec 16G / 32M
258 MB/sec 16G / 64M
260 MB/sec 16G / 128M
260 MB/sec 16G / 256M
259 MB/sec 16G / 512M
258 MB/sec 16G / 1G

Kernel 2.6.38.8:
352 MB/sec 16G / 1K
351 MB/sec 16G / 2K
352 MB/sec 16G / 4K
353 MB/sec 16G / 8K
353 MB/sec 16G / 16K
353 MB/sec 16G / 32K
353 MB/sec 16G / 64K
353 MB/sec 16G / 128K
351 MB/sec 16G / 256K
353 MB/sec 16G / 512K
353 MB/sec 16G / 1M
353 MB/sec 16G / 2M
353 MB/sec 16G / 4M
353 MB/sec 16G / 8M
344 MB/sec 16G / 16M
347 MB/sec 16G / 32M
353 MB/sec 16G / 64M
352 MB/sec 16G / 128M
352 MB/sec 16G / 256M
352 MB/sec 16G / 512M
351 MB/sec 16G / 1G

I can see a similar drop 205 MB/sec -> 100 MB/sec (starting from 256K block size) at my Non-SandyBridge SATA II netbook so the issue is probably not related to HW.

Any tips welcome.