KTD (10/9/2013)
I would suggest you find out your percentage of reads to writes before deciding this. The limited hardware you have listed here isn't going to support a very busy database but I have no idea what your dealing with in that area or even your disk speeds. . You could make better use of it if you discover you have 85%reads to writes for instance by going with raid 5.
The disks are 15K RPM SAS drives. The databases are used for analytics and have daily and hourly data loads from another database. It needs to be optimized for writes. So, RAID10.
KTD (10/9/2013)
Look what you are suggesting for TempDB and Log. That is going to get you at the most 160-200IOPS with 15K SAS drive.
Exactly!
But, as an aside, are you familiar with SQLIO? Here's my results on a two drive RAID1. I was expecting 200 IOPS max and I'm wondering why it reports what it does. I would have a tempdb with 8 files on the two drive RAID 1.
using system counter for latency timings, 2212939 counts per second
8 threads writing for 120 secs to file F:\TestFile.dat
using 64KB sequential IOs
enabling multiple I/Os per thread with 8 outstanding
buffering set to use hardware disk cache (but not file cache)
using current size: 24576 MB for file: F:\TestFile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 2583.37
MBs/sec: 161.46
KTD (10/9/2013)
A generic answer of separation may actually hurt you more in such a situation.
That was my thinking, but while researching this I almost always see the recommendation to separate them. Thus, my queries here to get some further insight.
It seems that you would support option A - aggregate the I/O over 7 disks mirrored to 7 disks and let the chips fall where they may? That was my original plan, but I started researching this some more to make sure it was the best plan, given the resources.
Is there any way I could test performance of option A over option C, in the next couple of days? Any tools I should try?
Thank you all very much for your comments!