As mentioned earlier, with RADOS classes, code is executed directly inside the OSD code base and so can harness the combined power of all of the OSD nodes. With a typical client application approach, where the client would have to read the object from the Ceph cluster, run computations on it, and then write it back, there is a large amount of round trip overheads. Using RADOS classes dramatically reduces the amount of round trips to and from OSDs, and also the available compute power is much higher than that single client could provide. Offloading operations directly to the OSDs therefore enables a single client to dramatically increase its processing rate.

A simple example of where RADOS classes could be used is where you need to calculate a hash of every object in a RADOS pool and store each objects hash as an attribute. Having a client perform this would highlight the bottlenecks and extra latency introduced by having the client perform these operations remotely from the cluster. With a RADOS class that contains the required code to read the object calculate the hash and store it as an attribute, all that the client would need to do is send the command to OSD to execute the RADOS class.