Hi Our service already run cassandra 1.0 on 1x ec2 instances(with ebs), and we saw lots of discussion talk about using ephemeral raid for better performance and consistent performance.
So we want to create new instance using 4 ephemeral raid0, and copy the data from ebs to finally replace the old instance and reduce some . we create the xlarge instance with -b '/dev/sdb=ephemeral0' -b '/dev/sdc=ephemeral1' -b '/dev/sdd=ephemeral2' -b '/dev/sde=ephemeral3', and use mdadm command like this mdadm --create /dev/md0 --level=0 -c256 --raid-devices=4 /dev/sdb /dev/sdc /dev/sdd /dev/sde after copying file and start the cassandra(same token as old instance it replaced). we saw the read is really fast always keep 2xxm/sec, but system load exceed 40, with high iowait, and lots of client get timeout result. We guess maybe it's the problem of ec2 instance, so we create another one with same setting to replace other machine ,but the result is same . Then we rollback to ebs with single disk ,read speed keeps at 1xmb/sec but system becomes well .(using ebs with 2 disks raid0 will keep at 2xmb/sec and higher iowait then single disk ,but still works) Is there anyone meet the same problem too ? or do we forget something to configure? thank you koji
