How can you optimize attachment handling for large binary payloads in Axiom?

Several levers combine to keep large-attachment workloads efficient rather than relying on any single setting doing all the work.

• Set an appropriate MTOM optimize threshold so binaries above a sensible size are shipped as XOP/MTOM attachments rather than inline base64, avoiding both the roughly one-third size inflation and the CPU cost of encoding/decoding.
• Stream from disk-backed sources - construct DataHandlers from a FileDataSource (or similar) rather than loading an entire large binary into a byte array first, so the attachment can be read and written incrementally instead of held whole in heap memory.
• Configure the attachments temp directory and caching thresholds so large incoming MTOM parts spill to disk rather than accumulating entirely in memory, particularly under concurrent load from many simultaneous large messages.
• Use serializeAndConsume() for outbound one-shot messages so the transport layer writes attachment bytes straight through from their source rather than buffering an extra intermediate copy.

These measures reinforce each other: a disk-backed DataHandler combined with a disk-backed attachments cache means a large binary payload can, in principle, pass from an inbound message straight through to an outbound one without ever being fully resident in memory at once - which is the practical ceiling on how large an attachment a given service can handle reliably under load.