Eric B. wrote: > Hi, > > I've been trying to read all the threads relating to which connector is > "best" to use HTTP (no SSL). I am planning to use Pound as an HTTP load > balancer in front of Tomcat as I have no need for all the bells and whistles > that Apache provides and Pound is fast and light. > > From what I've been reading through all the threads here is that the JIO > connector is the oldest and the most stable. The APR connector is basically > the same connector that is used in Apache httpd, so using the APR connector > would, in theory, give me the same performance as though I was using httpd. > Finally, the NIO is the latest addition to the Tomcat family that gives you > the benefits of a fully java non-blocking connector, and should perform > similarly to the APR connector for HTTP but be more sluggish on HTTPS. > Addtionally, given that NIO is the most recent, it doesn't have as much > "experience" as APR or JIO. > > That being said, I was leaning towards using the NIO connector for my > installation. However, I was pretty surprised and shocked when reading > "Tomcat - The Definitive Guide 2nd Edition" by Jason Brittain (O'Reilly > Press), that the JIO was the fastest and most responsive when service small > text files and 9k images. > (http://books.google.ca/books?id=vJttHyVF0SUC&pg=PA148&lpg=PA148&dq=tomcat+nio+advantage&source=bl&ots=i_8ssSnNf3&sig=MWgnLmZquhONBLLc5ivHQ6F61_Y&hl=en&ei=vwiDSsLgLMeltge4ianFCg&sa=X&oi=book_result&ct=result&resnum=4#v=onepage&q=tomcat%20nio%20advantage&f=false > > pp.138-148). In fact, their published benchmarks should that the JIO was > fastest, followed by APR, followed by NIO. Could that be attributed to > configuration parameters for the individual connectors? > > That seems pretty contrary to everything I've read on this list to date. > Can anyone shed some light on this descrepancy?
That just proves that one can construct a load test to show pretty much whatever you like. Your analysis above is pretty much spot on. The only thing I would add is that hard numbers are highly dependant on the environment and on your app. The only way to know for sure is to do lots of testing. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
