[Linux-PowerEdge] Storange power/performance throttling (by iDRAC or CMC?) on M620

Fumihito Yoshida fumihito.yoshida at gmail.com
Thu Jul 24 21:55:35 CDT 2014


Hi,

I met some problem of PEM620 storange performance limitation,
suspected module are iDRAC/CMC power control functions.

short summary:

- Sometimes, M620 falls unexpected "low power" mode, getpbinfo
  (at CMC cli) reorts very low budget(200-300W, expected 450-500W),
  and CPU power are limited.
- In this situation, shutdown system && reset iDRAC solve this
  "low power" mode.
- I suspect iDRAC/CMC bug, but I'm not sure root cause.

detail:

  somedays, getpbinfo returns below results, "Allocation" column=250W
  is notable point, other blade show about 400-500 W.

-----------------------------------------------------------------------------
[Server Module Power Allocation Table]
<Slot#> <Server Name>   <Power State>   <Allocation>    <Priority>
<Blade Type>
1       node05-01              250 W           1           PowerEdge M620
2       node05-02              250 W           1           PowerEdge M620
3       node05-03              250 W           1           PowerEdge M620
4       node05-04              461 W           1           PowerEdge M620
5       node05-05              496 W           1           PowerEdge M620
6       node05-06              496 W           1           PowerEdge M620
7       node05-07              250 W           1           PowerEdge M620
8       node05-08              479 W           1           PowerEdge M620
9       node05-09              250 W           1           PowerEdge M620
10      node05-10              250 W           1           PowerEdge M620
11      node05-11              250 W           1           PowerEdge M620
12      node05-12              479 W           1           PowerEdge M620
13      node05-13              496 W           1           PowerEdge M620
14      node05-14              479 W           1           PowerEdge M620
15      node05-15              250 W           1           PowerEdge M620
16      node05-16              461 W           1           PowerEdge M620
-----------------------------------------------------------------------------

  Slot# 1,2,3,7,9,10,11,15 are blade of interest. I test with
  CPU benchmark, but they can't get more power budget. Some
  blades get more budget and up to 450-500W levels, they are
  good.

  However, few nodes still in "constaint", their power state
  stay 250W. These "constraint" nodes have limited performance.

  We use Xeon E5-2670, but in this problematic situation, freq
  are limited under 2GHz (expected 3.1GHz in turbo state), and
  CPU C0:C1 balance are 20:80 ~ 1:99 (expected 100:0).

  I explore the solution, and I found poor workaruound, shutdown
  the system and reset iDRAC (racadm racreset), problem are
  definitely solve.

How to reproduce:

  Run blades for a long time(1-2 month?), typically repro about
  1-2 percentage(min, up to 10-20 percentage) units. In my case
  about 300 blades run 3 month, 4-5 blades(min) entered this
  symptom state.

  Yes, this is ineradicable, but poorly-reproducible.

Firmware/OS :

  - Scientific Linux 5 and 6
  - BIOS 2.0.19 + iDRAC7 1.46.45 (Build 4)

Anyone have solutions?



More information about the Linux-PowerEdge mailing list