Hey guys.<br><br>This is what I've been using Netty with:<br><br><a href="https://bitbucket.org/burtonator/peregrine/wiki/Home">https://bitbucket.org/burtonator/peregrine/wiki/Home</a><br><br>It's still alpha now and is just now running 10GB jobs. <br>
<br>Initially, I decided to just use Netty for the HTTP framework but ChannelBuffer sucked me in and I'm probably going to have all of our IO driven by ChannelBuffers... <br><br>There are just a bunch of performance benefits to doing it this way. <br>
<br>I also ended up building out some support for extending Map Reduce so that Keys and Values are now structs. This way you can do things like value.writeInt, value.writeDouble, etc. I will probably end up having these be facades on top of ChannelBuffer.<br>
<br>Hadoop uses Jetty for its HTTP support which means there's one thread per client. Some design decisions in Hadoop were pinned based on this and they weren't able to do some clever hacks that I can now do in Peregrine.<br>
<br>Kevin<br>