CODEWITHSUNDEEP
cassandradatastax-enterprisecassandra-3.0
root
root

Reputation: 3937

What it mean to be write heavy in cassandra?

I am kind of confused in terms of what it means to be write heavy in cassandra. Is this relative to the read frequency on a particular cluster or something absolute. e.g lets say in our environment if we have like 500 tps written and may be 100 queries per second. In this case write is relatively higher then the reads. But if we look at just the writes from cassandra perspective, is this really high ?

We have ssd in prod with around 64 gigs of ram.

Upvotes: 2

Views: 2252

Answers (1)

Chris Lohfink
Chris Lohfink

Reputation: 16400

Writes are much faster and the defaults are tuned around data models that write more than they read for denormalizing the data.

4k writes/sec per core kinda standard rule of thumb which is safe (per host). With big host like that if tuned correctly its perfectly reasonable to have ~100k-200k writes a sec (if dont run outta disk). When you introduce reads that drops significantly but its really a factor of data model and the client, it should be relatively easy to do 20k a sec with 50% reads/writes but it can definitely go higher. With a bad data model though its all out the window.

If your doing below 1k reads+writes a sec per host you shouldn't have many issues unless your data model is an anti pattern, and even then with that load it might work out.

Upvotes: 3

Related Questions