Large Capacity Drives verses Small Capacity drives
-
Would it be better to purchase a few large capacity drives or a large number of small capacity drives? I'm trying to come up with some idea on what would be a reasonable performing disk configuration. I'm currently thinking about doing the following
Large Capacity drive configuration:
2 mirrored os drives
2 mirrored tlog drives
3 raid5 data\index\backup drives
or
Small capacity Drive configuration:
2 mirrored os drives
2 mirrored tlog drives
6+/- raid5 data\index drives
10+/- raid5 backup drives
I know drive capacities are quite large now, but I seem to recall a large number of smaller capacity drives for database servers is better performance-wise. Your thoughts are greatly appreciated.
Gregory A. Larsen, MVP
-
generally the more spindles the better
Although using 16 drives oposed to 3 raises some questions.
Large drivers are generally more modern than smaller ones, and you get extra management with so many drives (space, improved controllers,...)
I'm new to constructing servers, why do you choose raid 5 for data instead of raid 1+0 (using 4 drives)? Because it will be mostly read?
-
Raid 5 was selected because of cost/GB Using Raid1 or 1+0 cost more since you have to buy 2 drives to get 1 drive worth of capacity. So basically if you have 4 drives Raid1 or 1+0 you only get to 2 drives of capacity, where as with 4 drives using raid5 you get 3 drives of capacity. So at least that is my understanding.
I'm thinking that possibly 16 small drives might have better throughput then 3 large drives.
Gregory A. Larsen, MVP
-
More drives = More Speed, if you've got the cash then RAID 10 is faster but you lose half your capacity, most systems we run here use RAID 5 and only the ones that need the extra performance use RAID10. Also go for the higher RPM drives (15,000rpm) if you can.
-
The more spindles the better <period> The only time when purchasing huge disks (146 Gb) pays off is in a SAN configuration where you can slice into LUNs and reslice the LUNs into Meta-LUNs. Of course you still ahve to pay attention to spindle distribution, again, the more the better.
RegardsRudy KomacsarSenior Database Administrator"Ave Caesar! - Morituri te salutamus."
-
Probably obvious and already considered, but if this is for a server array then you need to check how many disks you can physically fit in the server before going any further.
One of our client's "bright ideas" fell apart when this was pointed out to them.
-
"Raid 5 was selected because of cost/GB Using Raid1 or 1+0 cost more since you have to buy 2 drives to get 1 drive worth of capacity. "
If you want to save some money, use RAID-1 and do not buy anti-virus software (just joking).
Nowadays, the cost of the hardware is negliable compared to the cost of the OS, SQL Server, Virus Scanner and monitoring software. SQL Server Standard costs $5,737
per CPU and Enterprise Edition is $23,911 per CPU. So for a server with 4CPUs, Standard Edition will cost almost $23,000 dollars.
If you put 5 36Gb drives into a RAID-5, you get 144Gb of visible space. To get the same amount of visible space with RAID-1, you need 8 drives. As HP is selling Ultra320 15K Rpm 36.4GB drives for $300, you are only saving $900
You should also consider using 2 sets of drives for backups with transaction log backups to RAID-1 but database backups to RAID-5. I have seen situations where the transaction log fills because the log backups are io bound and cannot keep up.
SQL = Scarcely Qualifies as a Language
-
"...you are only saving $900..."
Yes, but Carl, some of us work for companies where $900 is a big thing. I recently specc'd out a couple of new servers and had to go with a smaller capacity RAID controller to save $200
--------------------
Colt 45 - the original point and click interface
Viewing 8 posts - 1 through 7 (of 7 total)
You must be logged in to reply to this topic. Login to reply