@Mikael Syska,

I would go with write policy: write through; and disk cache policy: disabled. This way- the data going through the PERC would always have to be committed to your disks before the controller acknowledges the write being completed. For reads, it will force the controller to go to disks instead of the controller cache; but depending on your SQL setting, you can have those reads be cached in your system memory instead of the controller cache. Btw- this is also how I set up things when I was performance benchmarking TPC-C and TPC-H in the good ol' days.

Hi,

This is a mixed environment ... this one raid1 is mosly used for SQL ... so would you still turn change the write policy to write throuhh?

Disk Cache Policy ... still not sure what this is ?

and Read Policy you would still leave at Adaptive read ahead ?

@Mikael,

I would go with write-through for SQL application b/c I always felt that it was better to have the IO committed to the disk rather than cached in the controller cache and being dependent on your controller battery to back up that controller cache.

Next, I think that disk cache policy refers to whether you want to use the controller cache to cache IOs. In the past, I've bypassed the controller cache to ensure that writes are committed directly to disk when they're marked as completed. As for the reads, they will have to come from the disks; but the server's memory should be able to cache your Read data without having to go to your controller cache if your queries are asking for similar data.

And I would leave the read policy at adaptive read ahead.

well ... I will try that ...

Think I will see if I can use the last 2 empty spaces in the Dell 2950 ... and then make that RAID1 a Pass-through disk to dedicated to the SQL server ... and then get 2x1.5TB for misc storage that dont need 15k disks.

Thanks for the input ...

@syska-

Let us know how it goes.

Regards,

Kong

Hi,

Just borrowed two disk ... a 1 TB 7200 rmp and a 450GB 15k rmp ...

Tested in the following ways with "HD Tune" ... which I now think sux ... no diff, så I think I will have to make a more real world test ... but how ... its way overkill to test move all the data and let it run for 4 days to seem any changes ...

Pass through with both.
online at the host and then created a fixed size harddrive image and connect them ( 90% of the original drive size )

The results were allmost the same ... bot the fixed image size hardrives won over the pass through disk ... off in my mind.

But it could also be because of the "HD Tune" ... dont know.

// ouT