I have been running a two node cluster on EC2 and for the past week or so my database writes have been totally bogged down. After some tests it looked like we’re hitting the disk I/0 bottleneck. To my surprise Disk I/0 was 5-6 times faster on the m1.large instance type.
I ran a cheap command to time the creation of a 1GB file. Here are the results. In both the situations the small test was run on /mnt which is considered to be a dedicated spindle.
On m1.small
[root@]# time dd if=/dev/zero of=testfile count=1 bs=1024M
1+0 records in
1+0 records out
real 0m11.298s
user 0m0.000s
sys 0m3.390s
On m1.large
[root@]# time dd if=/dev/zero of=testfile count=1 bs=1024M
1+0 records in
1+0 records out
real 0m2.982s
user 0m0.000s
sys 0m2.350s
Read about the various EC2 instance types here