This post is more than 5 years old
92 Posts
0
1405
Best Practice for Database / Storage Pool
Hi,
Usually when you have a OLTP database, best practice was to have separate spindles for DATA and REDO.
I use to do this with "the old fashion" RAID GROUP and MetaLUNs.One for DATA one for REDO aka 2 or more separate disks.
EMC says "Use one pool instead of raid group" well why not.. if performance is not the main subject .
Now with storage pool, I have only one pool and data/redo is spread on all spindles. So even if i create 1 or more LUN the data will be spread by 1G chunks per raig group defined inside my storage pool...
Does it means that now only one lun [thick or thin] is enough for DATA and REDO ?
I think that separate DATA and REDO on 2 separate lun is always a good thing in order to have more granularity in case of performance problems [LUN Migration] but i wish to have a clearly status/feedback.
Another question : I have heard that rebalancing data when you add disk in an existing storage pool is forecast for Q2 / Q3 for VNX BUT what about CX4 ?
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
March 3rd, 2011 19:00
but can you argue that if it's pretty much sequential write, it will be nicely cached and written as full stripe write with even less write penalty ?
kelleg
4.5K Posts
1
February 8th, 2011 14:00
I'll take a stab at the first question - from what I've seen here on the forum and in working cases, it is always best to separate the data abd redo (logs) on separate LUNs, even if in a pool. I would recommend the following on PowerLink - do a search with the keyword "leveraging"
White Paper- Leveraging EMC CLARiiON CX4 Virtual Provisioning for Database Deployments — Best Practices Planning.pdf
White Paper EMC CLARiiON Virtual Provisioning - Applied Technology.pdf
glen
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
1
February 8th, 2011 16:00
i would actually consider putting redo logs onto a traditional LUN/RG, if something happens to to the pool, redo logs and data file are gone all at once (chances are small but ...). By keeping redo logs on separate RG, you can always restore data files and replay redo logs to bring your DB up to date.
jps00
392 Posts
0
February 9th, 2011 05:00
I see merit in this advice for two reasons. First, the overall capacity of the Redo logs does not merit creating a separate pool for. Also, the I/O profile for redo logs (small-block, sequential) is not ideal for pools.
SKT2
1.3K Posts
0
March 3rd, 2011 11:00
I also happen to observe that the logs (such as exchange ) predicted to do sequential IO (more wr)are put into to RAID 10 instead of R5. So i was asking this to myself why R10 instead of R5 when IO is sequential. May be becasue of the vertical RG lay out with my R10 gives more protection to logs to use for recovery...
Storagesavvy
474 Posts
0
March 3rd, 2011 18:00
I would say the main reason for RAID10 vs RAID5 for logs is that, usually, logs are nearly 100% write. RAID10 has less penalty than RAID5 on writes and the recommendation is generally to move to RAID10 when Write % exceeds 30% of the total IO.
jps00
392 Posts
0
March 4th, 2011 05:00
Considering the usage or re-usage pattern of logs, RAID 5 is a better solution. Logging is not a performance sport.
RAID 5 offers a higher storage utilization than RAID 1/0. And, as the learned dynamox points out, the log sequential, small-block writes will be handled very efficiently though cache, assuming average cache activity.
kelleg
4.5K Posts
0
March 8th, 2011 12:00
Please remember to mark the question as Answered when you get the correct or best answer. Also, please award points to the person providing the best answer.
glen
SKT2
1.3K Posts
0
April 3rd, 2011 11:00
Thanks to all for sharing thier thoughts