How much latency does proxy chaining add vs having all features in a single proxy ?

[anilsr]

Understand that proxy chaining feature incurs no network hops, but interested in 1) how exactly the call is made and 2) what is the cost in latency ?

~~Q:S:TC~~

[richarst]

I would add it would be great to also see this compared to a shared flow

Inline vs proxy chain vs shared flow

[DChiesa]

yes, and tested under high load & concurrency!

I don't know the answer, but my feeling is that proxy chaining will be much faster than a proxy-to-proxy call because there is no need to serialize to a buffer, call out the the network layer, and incur all that delay. instead it is a call within the same process. There is some memory allocation, but much less than there would be with a network call.

So my feeling is: proxy chaining would be faster.

But we'd need to conduct in-depth tests in order to give a precise answer.

[#4]

@richarst, @Dino test it!. Share the results 😉

[DChiesa]

I'm hoping to be able to deliver a few data points on this, in the near future. I may have some opportunity to do comparative measurements at scale for a couple narrow scenarios, for proxy chaining and for shared flows.

[#6]

Anecdotal evidence is that proxy chaining is an order of magnitude faster for one of my use cases than service callouts or JSC based loop making an FQDN call to a similar chained endpoint. While I did not push this to failure, it stands to reason that the pressure on the message processor shifts from network overhead to CPU and memory with the chaining approach.

Your mileage may vary based on the specifics of the workload.