Arch Linux

LeCrayonVert

Member

Registered: 2010-09-01

Posts: 134

Slow Timing cached reads since Kernel 3.0

Hi !
Ever since I've upgraded to the new 3.0 kernel, I've got very slow timing cached reads with hdparm -T :

hdparm -T /dev/sda 

/dev/sda:
 Timing cached reads:   422 MB in  2.00 seconds = 211.01 MB/sec

I used to have much higher value there.
Can anyone confirm that ?

Thanks

---

I believe in a world I can and do understand. A rational universe, explained through rational means.

LeCrayonVert

Member

Registered: 2010-09-01

Posts: 134

Re: Slow Timing cached reads since Kernel 3.0

Ok I've found out the cause...
It seems that sata power management is triggered when on battery with laptop-mode. In /etc/laptop-mode/conf.d/intel-sata-powermgmt.conf, :

BATT_ACTIVATE_SATA_POWER=1

But the problem remains, after a certain period of inactivity (even on AC), the system decides to enable power management....

What is weird is that I've got :

 cat /sys/class/scsi_host/host*/link_power_management_policy
max_performance
max_performance
max_performance
max_performance
max_performance
max_performance

even though hdparm -T timing is wrong....

---

I believe in a world I can and do understand. A rational universe, explained through rational means.

1LordAnubis

Member

Registered: 2008-10-10

Posts: 253

Website

Re: Slow Timing cached reads since Kernel 3.0

I'll try it out later; I know I have this problem when suspending to disk... errr, hdparm -t goes down dramatically.

---

Any society that would give up a little liberty to gain a little security will deserve neither and lose both.
-Benjamin Franklin
The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man.
-George Bernard Shaw

LeCrayonVert

Member

Registered: 2010-09-01

Posts: 134

Re: Slow Timing cached reads since Kernel 3.0

Well, it seems pretty random, as you can see with a script I've made to measure every minute this timing :

22:19 Timing cached reads: 3952 MB in 2.00 seconds = 1978.15 MB/sec
22:20 Timing cached reads: 3608 MB in 2.00 seconds = 1805.90 MB/sec
22:22 Timing cached reads: 2812 MB in 2.00 seconds = 1407.03 MB/sec
22:23 Timing cached reads: 368 MB in 2.00 seconds = 183.98 MB/sec
22:24 Timing cached reads: 322 MB in 2.00 seconds = 160.94 MB/sec
22:25 Timing cached reads: 3864 MB in 2.00 seconds = 1934.35 MB/sec
22:26 Timing cached reads: 3758 MB in 2.00 seconds = 1881.55 MB/sec
22:27 Timing cached reads: 3820 MB in 2.00 seconds = 1912.05 MB/sec
22:28 Timing cached reads: 3816 MB in 2.00 seconds = 1910.26 MB/sec
22:29 Timing cached reads: 3246 MB in 2.00 seconds = 1624.30 MB/sec
22:31 Timing cached reads: 3368 MB in 2.00 seconds = 1685.49 MB/sec
22:32 Timing cached reads: 3718 MB in 2.00 seconds = 1861.45 MB/sec
22:33 Timing cached reads: 3130 MB in 2.00 seconds = 1566.35 MB/sec
22:34 Timing cached reads: 3818 MB in 2.00 seconds = 1911.58 MB/sec
22:35 Timing cached reads: 3766 MB in 2.00 seconds = 1885.55 MB/sec
22:36 Timing cached reads: 3738 MB in 2.00 seconds = 1871.52 MB/sec
22:37 Timing cached reads: 3004 MB in 2.00 seconds = 1503.01 MB/sec
22:39 Timing cached reads: 62 MB in 2.01 seconds = 30.88 MB/sec
22:40 Timing cached reads: 394 MB in 2.00 seconds = 196.91 MB/sec
22:41 Timing cached reads: 2 MB in 2.16 seconds = 950.19 kB/sec
22:42 Timing cached reads: 12 MB in 2.00 seconds = 6.00 MB/sec
22:43 Timing cached reads: 3918 MB in 2.00 seconds = 1961.64 MB/sec

And I'm almost certain that this behavior was also present in previouses kernels. I've noticed a long time ago that the "next track" function of rhythmbox is sometimes very slow.. I can confirm that it happens each time hdparm timing is high...

---

I believe in a world I can and do understand. A rational universe, explained through rational means.