cancel
Showing results for 
Search instead for 
Did you mean: 

I/O problems

Brad Scalio
Occasional Visitor

I/O problems

Greetings,

 

We have DL380G6 servers with a RAID configuration as follows:

 

Smart Array P410i in Slot 0
   Bus Interface: PCI
   Slot: 0
   Serial Number: 5001438008F7C420
   Cache Serial Number: PACCQ0E9VYQ05QX
   RAID 6 (ADG) Status: Disabled
   Controller Status: OK
   Chassis Slot: 
   Hardware Revision: Rev C
   Firmware Version: 3.00
   Rebuild Priority: Medium
   Expand Priority: Medium
   Surface Scan Delay: 3 secs
   Queue Depth: Automatic
   Monitor and Performance Delay: 60 min
   Elevator Sort: Enabled
   Degraded Performance Optimization: Disabled
   Inconsistency Repair Policy: Disabled
   Post Prompt Timeout: 15 secs
   Cache Board Present: True
   Cache Status: OK
   Accelerator Ratio: 100% Read / 0% Write
   Drive Write Cache: Enabled
   Total Cache Size: 256 MB
   Battery Pack Count: 0
   SATA NCQ Supported: True

 

 

My question is this ... first off, we enabled Drive Write Cache due to postGreSQL logging causing heavy I/O and disk wait at times of heavy loads.  But, the Accelerator Ratio is obviosuly set to maximize reads.  Since we don't have a BBWC (Pack Count - 0) - and yes we know that means the cache is lost on reboot or power outage - the questions is:

 

[1] Does the accelerator ratio matter?  Should it be 25/75 or 0/100 if we are only doing key scan reads.

[2] Can we not change the cacheratio with hpacucli or for that matter any RAID parameters without a BBWC battery.

 

I was debating there possibly being an issue with I/O reordering, using the CFQ scheduler, and a write cache, does the controller re-order writes again if the CFQ reorders them during times of heavy I/O?  Or, does this really even matter much since we are only mirroring two disks in RAID1, and scheduling/reordering really only comes into play when non-sequential writes across multiple spindles?

 

I appreciate any feedback.