AML
Is the Advanced Multithreaded Logger (pun intended). We saw about 60MB/s on one buffer box.
It’s transparent to the ATP’s - they are told a certain port number and won’t know the difference. No change in protocol or anything.
Right now, the ATP’s send buffers of events to the ET_SERVER. The et_server disassembles the buffers into events, and feeds them one by one into the pool.
The logger the grabs events, reassembles them into buffers, writes them out.
--> Overhead per event, not per buffer -- scales up to 25…30MB/s, not more.
The buffers from the ATP’s would be ready to be written out. No difference.
In comes the AML - instead of one server process per ATP, there is one server thread per ATP - the threads make the flow control to the output file easy, no dedicated logger any more. The pool just gets a fraction of the events to satisfy the monitoring needs.
That would, in the future, allow the ATP’s to send compressed buffers, you get compressed prdf’s right from the atp’s.