Reputation: 21
I've been testing Cassandra for some time on different hardware.
I'm experiencing a strange thing on the new machine; the writes are faster but the reads are two times slower then the old machine.
Does anyone have any idea why this could be happening?
Also if I try to run selects and updates in the same time, the selects are faster than the case where there are only select queries.
Upvotes: 2
Views: 1174
Reputation: 20210
What is the difference between the disks? Are they both 7200 RPM? and is the new machine much much bigger disks? The size of the disk can really slow down your reads due to the seek times. Most people run max 1T and hear some run 500G disks. We are planning on 1T disks but only using 500G with the playOrm open source project on top of cassandra (you can also optimize for reads by having a 1T disk but only writing on the inner circles so it's faster).
The other case sounds test case specific...is select perhaps reading the same updated values in which case it might get them from memory...not really sure there.
Upvotes: 2
Reputation: 168
Check if the garbage collection is taking more time. When you increased memory and if you are using the default settings for the JVM then your JVM size has increased and might be its spending more time doing garbage collection
Upvotes: 1
Reputation: 2316
I agree with Dean, you have test case specific row retrieves that might change your request results. Perform some retrieve that hit different part of your DB :)
SSD vs Hard disk performance are not to explain in cassandra. Prefer SSD 7200 RPM!!
Upvotes: 1