AIO Buffer & Executors Optimization Update

[clebert.suconic]

I have the changes done at my workspace already.

I could get a successful run if using r6654 (before the pool refactoring), and a few failures if using 6653 (same ones as I have seen on hudson).

After these changes, buffer creation is going to be pretty fast and buffer reuse will be fast.

I will also be using executors for thread pools. I will be using a single-thread executor for writing (as it is done at the moment, but the executor will be shared among all the files), and another one to reuse threads on pollers.

I'm not using the thread-pool from MessagingServer because:

I - The poller thread will be in use as long as the file is opened
II - It would be really dangerous sharing the executor with MessagingServer. I really need a thread available for polling data on AIO.



I will commit it tomorrow morning. (I just didn't want to commit now with the testsuite broken as it is now).

[clebert.suconic]

BTW: What happened to perfSender and perfListener?

We were using those to measure the release on Alpha. I was using those to measure AIO.

[timfox]

BTW: What happened to perfSender and perfListener?

We were using those to measure the release on Alpha. I was using those to measure AIO.

PerfSender and PerfListener used the old example framework so could not work as is with the new one.

You could port them to the new framework.

[ataylor]

PerfSender and PerfListener used the old example framework so could not work as is with the new one.

You could port them to the new framework.

I had this as a todo. Typically the examples start the server as part of the examples. For this i was going to have a separate target since the producers/consumers can be started in different orders. so something like,

ant runServer;
ant runConsumer;
ant runProducer;

Clebert, feel free to port them.