Are multiple AMI instances run on the same physical servers? If so, does the load from one customer's server instance negatively effect all of the others on the same hardware? I understand that you can instantly spool up additional servers to dynamically handle load, but do those individual instances suffer from performance issues if they're loaded on busy hardware?
1) yes. 2) no. 3) no. ...there is a minimum guaranteed performance level for each instance type. for small instances (1 core), each user gets a minimum of 1/N of the system resources. If you fire up an extra-large instance, you essentially get the whole box to yourself, including I/O.
Thanks for the answers folks. My roommate is talking about needing tens of thousands of dollars of servers for a company he's thinking about starting and and my immediate thought was "why?" I just got my first EC2 instance up and running from the terminal in OS X so I'm doing my little happy dance. Thanks for the help.
yeah go with EC2 first... get users... and more users... until EC2 doesn't cut it... then turn the virtual machines into dedicated servers somewhere. Truth is you'd have to spend tens of thousands of dollars to get the reliability amazon should be selling you for much less.
The only downside to them, in my opinion, is that disk access is not super fast. This makes running databases on them much slower than a properly set up real machine.
That said, I fully expect amazon to fix this, just as they have already added high memory and high cpu instances. They seem to take feedback seriously and have addressed most of the major complaints to date.
Being able to provision new machines with a simple command is reason enough to use them over more traditional places.
we just ran some extensive testing to move a highly trafficked website from a cluster of dedicated servers (at theplanet) to ec2. note that our application is to use ec2 as a real-time web server. the bottom line:
dedicated servers allow for much faster ip takeover (critical for high-availability) and had lower latency. ec2 rocks for scalability.
in the end we decided to hold off on ec2 until they address these 2 deficiencies.
see: http://docs.amazonwebservices.com/AWSEC2/2008-02-01/Develope...