I recently attended the UK Couchbase conference and while discussing hardware requirements there was a specific mention on avoiding the use of VM’s and SAN storage.
Unfortunately in our environments that’s all we currently have which is not a great start
VM’s are not cloud based.
Anyway my real question is why do you make these suggestions, in order to get some decent hardware to run Couchbase on in our production environment I kind of need a technical answer as to why this is the case. Our storage experts beleive our SAN should out perform local disk because of the number of spindles in our array.
I’m not a storage guy or hardware techie so really could do with some advice on this subject.
Thank you again for your response, this puts my mind at rest.
The VM’s allocated are all on different chassis and have multiple links to the SAN. The SAN is pretty robust apparently
My main concern here would be the latency of the SAN, apparently it is great for high I/O but the latency is not as good as a local disk. As mentioned I’m no storage expert just relaying information from our storage guys.
I am not sure what was the exact message you got during Couchbase [UK], but I know many projects that are running Couchbase on VMs (cloud or not), but you have to be sure that your system is “safe”:
- do not put multiple nodes of your cluster on the same hardware: if you lose this hardware you will lose “multiple” nodes of your database and chances are Couchbase can not get the data from replicas during failover
- SAN: same reason need to be sure your do not lose data, and also have good performance to do the I/O (data access , but more importantly all the compaction process)