July 10, 2008 at 4:47 am
Hi All
I have created a new SQL server Machine, this machine has got all the drives on RAID 5, and i have made 3 drives, one for OS one for data and the other for log, each has got its own LUN, the Config is as follows
C (OS) -- RAID5_1
D (Data) -- RAID5_2
E (Log) -- RAID5_3
Will this give me a better performance rather than having all the 3 drives on the Same LUN
Please Suggest
I am running SQL 2005 on virtual environment
Cheers
Sujith
July 13, 2008 at 7:08 pm
Are these real LUNs backed by seperate physical sets of disks? If so, yes they should give better performance than having them all on one LUN. However RAID5 is generally a poor choice for most volumes - the write performance tends to be horrible in comparison to RAID1/RAID10 unless you have a massive write cache dedicated to it.
Typically most servers use RAID1 for the OS, either RAID1 or RAID10 for the Logs as they are write-intensive, and RAID10 for the Data volumes.
Regards,
Jacob
July 13, 2008 at 7:39 pm
Second Jacob's advice. Especially in a virtual environment, be sure that your LUNs are physically separate and try to separate the storage paths as much as possible.
also, don't allow your VMs to float.
July 14, 2008 at 12:12 pm
Thanks Jacob / Steve
Yes this are Lun's with seperate disks, and i have tried RAID1 earlier on VMWare for data and OS files, it gave me a bit low performance than RAID5, its strange, since every article i read suggested me RAID 1, i just have migrated my production machines to the above said configuration, performance seems to be improved by a percent, this is not huge, but it dint drop, i am waiting for 2 more days to see the actuall performance during peak and off pear periods,
Thanks for the posts again
Cheers
Sujith
July 14, 2008 at 5:22 pm
No problems Sujith,
You can get good read performance from RAID5, so if you compared RAID1 (2 spindles only) to RAID5 (3+ spindles) you would likely have seen improved reads (the more spindles the bigger difference). However you've got an increased risk (more spindles with only 1 disk redundancy means more chance of data loss) and your write performance will suck. Still, if your workload has a very high read/write ratio (eg a reporting DB) it might be fine.
RAID5 doesn't make sense on the OS (waste of disks) or Logs (needs good write perf) though.
Regards,
Jacob
July 15, 2008 at 10:36 am
i tought of that as well Jacob, i am using a HP server, there is a monitoring system used by HP Servers and i am having a warm spare on the server, that will be used once a disk has been failed,
As u said about the Log file, i am thinking to change to a diffrent LUN with RAID1, and check the performance on that
Thanks again.
August 1, 2008 at 1:14 pm
Don't forget about tempdb! Place tempdb data files in their own independent physical array, if possible. This may be more critical than RAID level.
__________________________________________________________________________________
SQL Server 2016 Columnstore Index Enhancements - System Views for Disk-Based Tables[/url]
Persisting SQL Server Index-Usage Statistics with MERGE[/url]
Turbocharge Your Database Maintenance With Service Broker: Part 2[/url]
Viewing 7 posts - 1 through 6 (of 6 total)
You must be logged in to reply to this topic. Login to reply