Is RAID 1+0 necessary?
We are using Vertica 7.2 community version, 300GB of data on one node.
HPO recommends RAID 1+0.
https://community.dev.hpe.com/t5/Vertica-Knowledge-Base/RAID-Storage-for-HPE-Vertica/ta-p/234225
But I do not envision a problem with write performance which is the point of RAID 0. I am not planning to load, insert or write large amounts of data. I am planning to *query* large amounts of data, but not write a lot. It seems to me that RAID 1 is sufficient.
Is there anything wrong with using RAID 1 instead of 1+0?
0
Comments
With the long term perspective, I believe RAID 10 will do better as hard drives might fail. RAID 10 leaves space for one hard drive in the stripe (where segmented data is present) and the mirror (where replicated data is present) failing and still one can have all the data.
Thought I'd add something here about RAID in case anyone reading this is considering Vertica on AWS...
HPE's document, Using the Vertica Analytics Platform on AWS, on p. 14, points out that...
.."For optimal performance with EBS volumes, Amazon recommends that you configure them in a RAID 0 array on each node in your cluster".
...with a link to the AWS docs on
http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/raid-config.html
which states...
"When I/O performance is more important than fault tolerance; for example, as in a heavily used database (where data replication is already set up separately)."
What kind of and how many storage devices are present on each node (HDD SATA 7200 or HDD SAS 10K or SSD SATA etc) ? What kind of controller is used?
RAID level discussions without knowing hardware config details are rather useless.
If there are at least 6 storage devices and you're mostly concerned about read performance, you can choose
for data storage (vertica TEMP could be on a separate device)
- several RAID1 arrays (as Vertica supports multiple storage locations per node)
- 1 RAID6 array (4+2 or 8+2 drives are often used); could be very fast for reads with good controller
- 1 RAID10 array.