pchalamet
commented
11 years ago Yes this uses to exist in previous version... Buffering should be done, I do agree. Go on then !
- Pierre
-----Original Message----- From: "Alex Haslehurst" notifications@github.com Sent: 10/07/2013 11:35 To: "pchalamet/cassandra-sharp" cassandra-sharp@noreply.github.com Subject: [cassandra-sharp] Performance Improvement (#62)
Currently, using the POCO interface ReadByteArray() is called on SocketReadOnlyStream for every column and every row of the results of a query. In my case that's 4.2m times for a typical query. This seems to hammer the CPU. Probably because of all the Windows socket API calls? The only solution I can see is to buffer each frame body into a memory stream. I've forked and stripped your driver right down and just by implementing buffered frames, see ~50% better performance for large queries. Maybe in cassandra-sharp a BufferingFrameReader could work? Such an implementation could provide an easier way of supporting compressed frames also. — Reply to this email directly or view it on GitHub.
axle-h
Currently, using the POCO interface ReadByteArray() is called on SocketReadOnlyStream for every column and every row of the results of a query. In my case that's 4.2m times for a typical query. This seems to hammer the CPU. Probably because of all the Windows socket API calls? The only solution I can see is to buffer each frame body into a memory stream. I've forked and stripped your driver right down and just by implementing buffered frames, see ~50% better performance for large queries. Maybe in cassandra-sharp a BufferingFrameReader could work? Such an implementation could provide an easier way of supporting compressed frames also.