Reduce disk IO or do not use disk at all

Hello,

Your data need to come from "somewhere" if they are in Couchbase they will be on disk, BUT they will then be all loaded in memory of you have enough RAM on your cluster.
Once they are in memory the IO/Disk activity will be very very limited (only when you update/delete entries)

If your data are coming from another place you can use a Memcahe bucket and use Couchbase as a simple caching layer.

Could you please give us more information about your dataset how big, type of operations, lifecycle, ...

Thanks

Thanks for your reply.

We are using Couchbase DB for storing real time data. All of our data has TTL expiry of 1 day or less. As we will have multiple nodes we don't require disk persistence to these nodes.

We will have very high Inserts/Updates with ratio of 9:1. No deletes commands as we use TTL expiry.

One e.g. of type of data is user's session. Size of this DB could grow to few hundred GBs.

We cannot have memcached bucket as we need this cache to replicate to multiple nodes & datacenters.
We are also using features that Couchbase provides so memcached is not an option.

As we do not require persistence it would be good to know if it could be disabled. Even if we could disable certain writes to disk it would be good. For e.g. if we could disable views/indexes or logs/stats to be written to disk. As we have read other threads of this forum it seems like disabling disk writes is not possible.

So my question is what are the ways we may reduce disk IO requirement?

As someone in this form said, a not-so-good way of reducing disk IO would be store data in tmpfs or ram disk. Is there any better way?