From Wikitech
Location: pmtpa
Node name: storage1.pmtpa.wmnet (fingerprint)
Serial: SM27965
Usage: apache
This device has been decommissioned.
Icinga?: host status services status


  • 2007-09-19: Replaced disk 13.
  • 2007-03-15: Replacement server arrives from SM. Racked and loading FC4 on it to test that it works.
  • 2007-02-22: Attempted to do the HDD tests required by SM to replace server due to HDD slowdowns. Emailed SM to get the entire thing replaced, there have been too many issues.
  • 2007-02-19: Replaced PDUs and Mainboard for the issue of the system not wanting to reboot without pulling all the power and plugging it back in.
    • Attempting to load Ubuntu shows the system experiencing the HDD throttling issue it first had out of the box. An email has been sent to SM about this on 02/17/07.
  • 2007-02-09: Loading Ubuntu.

Concerning the slow raid:


We will track this issue under Ticket: #15298.

It looks like the most recent storage server purchase you made was under 
order #S2025934 [SM24266, SM24267] for 2 x K300s.

We may have seen the issue you are describing to us before. The internal 
host drives have a feature where if there is too much vibration in the 
server the write speeds will slow way down in order to prevent damage to 
the platters. With these storage servers you would notice this issue if 
the server rack is vibrating more than usual or if the exhaust fans in 
the rear of the chassis are causing too much vibration. I would like to 
target the fans first.

If you are able to, please install lm_sensors on one of the storage 
servers and send us the fan speed output. This will allow us to narrow 
it down to which version of the exhaust fans your servers have in the 
rear of the chassis.

Thanks in advance,

Jonathan McQuistion
Silicon Mechanics Support
Tests:    (All outcomes are the result of two tests averaged together.
 Tests done with FC5 CD in rescue mode. Raid controller in question is a 3ware 8006 with Raid 1 on 2 80GB drives)

W/O Exhaust Fans and no raid fans (Only CPU Fans)
   dd if=/dev/zero of=test bs=1024 count=10000000   <-- 106MB/s
   dd if=test of=/dev/null                          <-- 241MB/
W/O Exhaust Fans -  (3 Raid fans and CPU fans)
   dd if=/dev/zero of=test bs=1024 count=10000000   <-- 98MB/s
   dd if=test of=/dev/null                          <-- 223MB/s
W/ Exhaust Fans - (All fans on)
   dd if=/dev/zero of=test bs=1024 count=10000000   <-- 81MB/s
   dd if=test of=/dev/null                          <-- 252MB/s
With Exhaust fans and Storage 2 powered on. (Its racked below Storage1)
   dd if=/dev/zero of=test bs=1024 count=10000000   <-- 72MB/s
   dd if=test of=/dev/null                          <-- 237MB/s

If there is caching taking place, is consistent. (Whatever is cached with the fans off, 
the same thing is cached with the fans on, and its slower)

There are definitely slower writes taking place on this raid controller with more fans. 
Obviously we cannot function without fans.