Example of Cluster SLOWER than InnoDB -- help?

[John Muehlhausen]

Hi all,

I am inserting a message into a Cluster table and then sending a notification to the receiver (if any). The receiver then reads and deletes the message record.

If there is no receiver then a single node Cluster (on a Powerbook G4 / OS X) is the same speed at inserting as InnoDB -- about 100 transactions per second (which seems slow to me).

If there is a receiver then InnoDB performance remains good but the Cluster performance slows to 10 messages per second!! :-(

Here is basically what is happening:

put:

set autocommit = 0
select put_seqnum from <queueName>_state for update
update <queueName>_state set put_seqnum = <next>
replace into <queueName>_messages (seqnum, message) values (..., ...)
update <queueName>_state set dirty = 1
commit

get:

set autocommit = 1
select dirty from <queueName>_state
// if not dirty then wait (with timeout) for notification
select get_seqnum from <queueName>_state
select seqnum, message from <queueName>_messages where seqnum > ... order by seqnum
loop
{
set autocommit = 0
update <queueName>_state set get_seqnum=...
delete from <queueName>_messages where seqnum=...
commit
}
set autocommit = 1
update <queueName>_state set dirty = if( put_seqnum > get_seqnum, 1, 0)

Ideas?

Thanks,
John