If your cache is near-perfect and all your content is served from memory, you'll see hardly any read operations on your storage system.
Since write operations won't be cached to any large degree, your systems does need to write those to disk to make them persistent, those will then quite quickly dominate your IO.
That seems to be the case in your diagram. That does not necessarily need to be a problem though, it doens't look like write IO is a real bottleneck yet.
As to why you get write operations on a server:
- Unless you use specific tuning options your file system will record the last access time (
atime
) for each and every file access. That can add up. Use the noatime
or relatime
mount options to reduce that.
- You probably collect logs as well.