What is the best threading model for 3k symbol subscription + low latency

Hi,

We have a single-threaded feed application (same as rmdstestclient) that is currently experiencing huge latency when subscribing to ~3k Shenzhen symbols.

We were looking into the possibility that our app is a slow consumer, and that we may need to perform some changes on our end.

We believe that the single-threaded model should be ok for the said amount of subscriptions, but is there any recommendation on what threading model should be used per market?

Is there any difference between the Shenzhen market and the other markets in terms of:

Threading configuration required?
Volume of Data?
Manner by which the data was sent?

Thank you very much!

Find more posts tagged with

c++

trep

rfa

rfa-api

multi-threading

32-bit

rmds-test-client

Accepted answers

umer.nalla

Hi @exchange-notifications

The RFA APIs are content-neutral and we don't generally recommend one threading model over another for specific exchanges or asset types etc.

It would be difficult for us to recommend a specific threading configuration without performing extensive testing using your 3k symbols, hardware and network, what processing you do with the payload once received etc - as all these factors can affect the performance and would be outside the scope of this Q&A Forum.

Generally speaking the threadModel : Single is normally recommended for lower latency - vs Dual for higher throughput.

Do you see the same latency issue when testing with rmdstestclient?
Does the latency with your app gradually increase as you run the application for longer? i.e. does it start off ok - but as time progresses, the latency increases?
What kind of latency figures are you seeing? Do these change during the trading day?
What are the typical update rates you are seeing?
Have you tried testing with Dual threadmodel? What impact do you see on latency?
Are your ADH and ADS configured for low latency? If not, I would recommend raising a MyRefinitiv support ticket to obtain some assistance in this area. I did find a TREP Tuning Guide on My.Refinitiv which you may find useful.

All comments

umer.nalla

Hi @exchange-notifications

The RFA APIs are content-neutral and we don't generally recommend one threading model over another for specific exchanges or asset types etc.

Generally speaking the threadModel : Single is normally recommended for lower latency - vs Dual for higher throughput.

Do you see the same latency issue when testing with rmdstestclient?
Does the latency with your app gradually increase as you run the application for longer? i.e. does it start off ok - but as time progresses, the latency increases?
What kind of latency figures are you seeing? Do these change during the trading day?
What are the typical update rates you are seeing?
Have you tried testing with Dual threadmodel? What impact do you see on latency?
Are your ADH and ADS configured for low latency? If not, I would recommend raising a MyRefinitiv support ticket to obtain some assistance in this area. I did find a TREP Tuning Guide on My.Refinitiv which you may find useful.

EXPLORE OUR SITES