12h - Should R & W cache be disabled?

[Michael Treadwell_1]

Output from arraydsp -a. It shows read & write cache disabled. Does this impact performance and should it be enabled?


Vendor ID = HP
Product ID = C5447A
Array serial number = 0000000F5476
----------------------------------------------------
Overall State of Array = READY
Array configuration:
Active Hot Spare Desired = ENABLED
Auto Include = ENABLED
Auto Rebuild = ENABLED
Rebuild Priority = HIGH
Capacity Depletion Threshold = 0%
Write Working Set Interval = 8640 seconds
Language = ENGLISH
Log Full Warning = DISABLED
Volume Set Partitioning = DISABLED
Format Pattern Fill = DISABLED
Subsystem Type ID = 12
LUN Creation Limit = 8
Maximum LUN Creation Limit = 32
Array SCSI configuration:
Controller X SCSI Address = 0
Controller Y SCSI Address = 1
Write Cache = DISABLED <-
Read Cache = DISABLED <- SCSI Parity Checking = ENABLED
SDTR = ENABLED
WDTR = ENABLED
Terminator Power = ENABLED
Unit Attention = ENABLED
Disable Remote Reset = ENABLED
Secondary Controller Offline = DISABLED
Very Early Busy = DISABLED
Queue Full Threshold = 1952

[A. Clay Stephenson]

Do a man of arraycfg. You will find that cache is always enabled. You are only able to change what is DISPLAYED to the host.

[Steven Sim Kok Leong]

Hi,

We had an XP256 with 6 GB of cache. Cache does matter a lot because instead of reading from the slower disks with moving mechanical parts, the cache is read for frequently-accessed data. This dramatically improves performance.

Thus, I do not think that the disk array cache should be disabled.

Hope this helps. Regards.

Steven Sim Kok Leong

[S.K. Chan]

Attached is a doc that I used to have, it explains how the "write cache" flag is related to the "Resiliency Setting".

Hope it helps ..

[Michael Treadwell_1]

Clay,
Thanks. I saw what you mean under 'man arraymgr'. If read cache is always enabled, why would you not want to report it to the os? Write cache I can understand depending on what behavior you desire.
Personally I would think you'd want both to be reported. Is there a good reason not to?

[MANOJ SRIVASTAVA]

Hi Michael

Essentially the cache should be enabled , we have a mixture of the arrays , EMC , and whenever we hard reset them it always take like 2-3 days for the array to start performing properly that always is because the cache is empty when the array starts fuctioning again. This also will definaltey effect performance considering any databse env has lots of r/w done .


Manoj Srivastava

[Bill McNAMARA_1]

well in all honesty they do nothing... they should be disabled in a windows environment, an os that doesn't seem to like the option... don't ask me why.

In any case, you are caching in unix whether it says you are or not.
Can't rememeber where I read that, think it was in the arraymgr man.

Later,
Bill

[A. Clay Stephenson · Accepted solution]

Hi again Michael:

I had to dig through my notes but I did test the setting several years ago (in HP-UX 10.20)
and found no measurable difference in performance of either setting of the read or write cache - again, it really is enabled regardless of this setting - this just changes the appearance to the outside world.

In that same section of my notes, I also found a note to myself that HP reccomends cache DISABLED for NT. I have never tested this in the NT world but in HP-UX it makes no difference though I will admit I have not tested it in HP-UX 11.x. That might be a project for you.

[Michael Treadwell_1]

Thanks, people!

The general consensus is the settings will have no effect.