SATA read speed gets worse in linux 3.14.52

[richardhu]

Hello, all:

I test SATA performace on our custom board with intel SSD and figure out performance difference in kernel version.

My SSD is INTEL SSDSA2CT040G3.

The benchmark under windows 7 64 bit with CrystalDiskMark 5.1.2:

The test commad i use on target board:

echo performance >  /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

Write test:

echo 3 > /proc/sys/vm/drop_caches

dd if=/dev/zero of=tempfile bs=1M count=1024 conv=fsync

Read test:

echo 3 > /proc/sys/vm/drop_caches

dd if=tempfile of=/dev/null bs=1M count=1024

Test Result:

(Rootfs is buildroot)

linux 3.14.52:

Write:

1073741824 bytes (1.0GB) copied, 26.856296 seconds, 38.1MB/s

Read:

1073741824 bytes (1.0GB) copied, 13.097736 seconds, 78.2MB/s

linux 3.10.53:

Write:

1073741824 bytes (1.0GB) copied, 28.990223 seconds, 35.3MB/s

Read:

1073741824 bytes (1.0GB) copied, 8.182314 seconds, 125.1MB/s

linux 3.0.35:

Write:

1073741824 bytes (1.0GB) copied, 36.756555 seconds, 27.9MB/s

Read:

1073741824 bytes (1.0GB) copied, 10.264888 seconds, 99.8MB/s

As you see, the read speed is 30% worse in linux 3.14.52 then linux 3.10.53.

Any idea ?

BR,

Richard

[igorpadykov]

Hi richardhu

one can check who is holding ddr bandwidth using profiling tool.

There is a yocto project recipe imx-test that provides a Multi-Mode DDR Controller

profiling tool mxc_oprofile_test: : run it  -  /unit_tests/mmdc2 <master>: where master is

described in Table 44-10. i.MX 6Dual/6Quad AXI ID  i.MX6 RM

Best regards

igor

-----------------------------------------------------------------------------------------------------------------------

Note: If this post answers your question, please click the Correct Answer button. Thank you!

-----------------------------------------------------------------------------------------------------------------------

[richardhu]

Hello, igor:

Thanks for your reply.

I run the "mmdc2" test as follows:

In linux 3.14.52:

=========================================================

root@edm-fairy-imx6:~# /unit_tests/mmdc2 SUM

i.MX6Q detected.

MMDC SUM

MMDC new Profiling results:

***********************

Measure time: 501ms

Total cycles count: 264084408

Busy cycles count: 69431977

Read accesses count: 873416

Write accesses count: 1657

Read bytes count: 55592984

Write bytes count: 53024

Avg. Read burst size: 63

Avg. Write burst size: 32

Read: 105.82 MB/s /  Write: 0.10 MB/s  Total: 105.92 MB/s

Utilization: 5%

Overall Bus Load: 26%

Bytes Access: 63

root@edm-fairy-imx6:~# /unit_tests/mmdc2 DSP1

i.MX6Q detected.

MMDC DSP1

MMDC new Profiling results:

***********************

Measure time: 500ms

Total cycles count: 264083472

Busy cycles count: 69405646

Read accesses count: 864288

Write accesses count: 0

Read bytes count: 55314432

Write bytes count: 0

Avg. Read burst size: 64

Avg. Write burst size: 0

Read: 105.50 MB/s /  Write: 0.00 MB/s  Total: 105.50 MB/s

Utilization: 4%

Overall Bus Load: 26%

Bytes Access: 64

=========================================================

In linux 3.10.53:

=========================================================

root@edm-fairy-imx6:~# /unit_tests/mmdc2 SUM

i.MX6Q detected.

MMDC SUM

MMDC new Profiling results:

***********************

Measure time: 1000ms

Total cycles count: 528087080

Busy cycles count: 138780500

Read accesses count: 1730322

Write accesses count: 498

Read bytes count: 110729160

Write bytes count: 15936

Avg. Read burst size: 63

Avg. Write burst size: 32

Read: 105.60 MB/s /  Write: 0.02 MB/s  Total: 105.61 MB/s

Utilization: 4%

Overall Bus Load: 26%

Bytes Access: 63

root@edm-fairy-imx6:~# /unit_tests/mmdc2 DSP1

i.MX6Q detected.

MMDC DSP1

MMDC new Profiling results:

***********************

Measure time: 1001ms

Total cycles count: 528091376

Busy cycles count: 138768053

Read accesses count: 1728000

Write accesses count: 0

Read bytes count: 110592000

Write bytes count: 0

Avg. Read burst size: 64

Avg. Write burst size: 0

Read: 105.36 MB/s /  Write: 0.00 MB/s  Total: 105.36 MB/s

Utilization: 4%

Overall Bus Load: 26%

Bytes Access: 64

=========================================================

As test result, the DDR bandwidth is occupied by DSP1 in both linux 3.10.53 and linux 3.14.52.

So, it seems it's not the root cause to result in bad performance of SATA.

How can I fix this issue?

Thank you !

BR,

Richard