Talk:GameZone/Performance on multiple servers
CS290F Fall 2006 - UCSB Computer Science - Thorsten von Eicken
Good work guys, looks like you've done a lot of work.
Comments:
- What are the different lines on these graphs? AllInOne, OneApp, TwoApp, ThreeApp are not good labels. Write something more informative that this in the legend, and discuss it somewhere at the top of the page so that the reader knows what you are doing!
- The rate per mongrel graph is not very interesting. Most people have found that varying the number of mongrel instances started by mongrel_cluster doesn't have much impact beyond threee or four.
- It would be much more interesting to look at how the performance scales when you ramp up the number of EC2 instances that are serving.
Stefan 12:18, 7 December 2006 (PST)
Hi Stefan,
I will gladly change the tags on the graphs, but I would like to refute your second comment. What I am trying to find in the mongrel instances graph is not the impact of the number of mongrel servers per machine, but the effective reply per second per mongrel instance as we increase the number of app servers. This helps because :
- It gives me information that the reply/sec per mongrel server is nearly constant, therefore, we are experiencing linear scaling.
- . It is clear that as we are increasing the number of instances, each mongrel instance is still serving the same number of pages. This tells us that the DB is not overloaded. Had it been overloaded, we would have seen a dip in the number of reply/second per mongrel.
IMHO, it is this graph that will give out most clearly, if my bottleneck is the database or not.
--Keshava 20:54, 8 December 2006 (PST)
Good point. It's certainly not harmful to have this information :)
