Hi!
I’m still about to implement my first kdb gateway service and have some little concerns.
Suppose I’m writing gateway (using deferred response, or async gateway - irrelevant to question). And suppose I received response from leaf service and need to pass it back to caller.
This will incur otherwise unnecessary deserialization/serialization step. If data is complex (large tree, for example) - this is lots of unnecessary allocations/deallocations.
Only thing I can imagine is to serialize data on leaf service using -8! and decompress it closer to consumer, when necessary. This way, overhead is closer to two large extra memcpy calls (from network receive buffer to kdb heap, and same on the way out).
Any other ways to manage it? Or serialization/deserialization CPU overhead is not that big?
Regards,
Sanny.