Slowness with large payloads

When reading a proto off the wire, Netty Passes the chunks of data read up to the message deframer, which stores them in a composite byte buffer. It passes this composite as an InputStream to CodedInputStream for decoding, which itself makes many copies.

It seems like it would be more efficient to pre allocate a buffer of the appropriate size (since we know the message length) and concat to it each chunk that netty passes up. This would be a copy, but it would make it so that CIS doesn't have to copy in its inefficient manner.

This would also free up the buffers Netty uses to read chunks of data off of the wire more quickly.

Discovered when trying to max out a very high speed network link.

Inizia tracciando come Netty passa i wire chunks al message deframer e come il composite buffer raggiunge CodedInputStream. Esegui benchmark delle letture di payload di grandi dimensioni e confronta la copia dei buffer e il throughput; il lavoro è considerato completato quando la copia è ridotta e le prestazioni migliorate senza modificare il comportamento della decodifica.

Scritto dal modello di indicizzazione a partire dal testo della issue.