This is specific to a consumer app written using EMA c++ library.
What if the event handling code (i.e. code that handles refresh/update/status) messages is too slow to keep up with the tick rate?
Does ema has a internal queue where ticks are buffered or it directly invokes handler from the reader thread? Is there a way to configure?
Or the ticks would just be dropped (temporarily or permanently?).
Thank you for your participation in the forum. Is the reply below satisfactory in resolving your query? If yes please click the 'Accept' text next to the reply. This will guide all community members who have a similar question. Otherwise please post again offering further insight into your question. Thanks, AHS
Please be informed that a reply has been verified as correct in answering the question, and has been marked as such. Thanks, AHS
In EMA, there is no internal queue where ticks are buffered. It uses the application thread (User Dispatch mode) or EMA thread (API Dispatch mode) to dispatch received messages. If the application delays in dispatching messages, it can result in slow consumer behavior.
When the consumer is slow, TCP receive buffer of the consumer side will be full. Therefore, the provider side can't send the data to the consumer.
The provider side has output buffers which hold the data before sending to the network. When the consumer is slow, the output buffers of the provide side will be depleted because it is unable to flush the data in the buffers to the network.
When the provider can't get the output buffer to write the data, it can decide what to do in this situation. For example, ADS (Advanced Distribution Server) will cut the connection. However, if Traffic Management is enabled in ADS, ADS will aggregate the data when output threshold is breach. Then, it will send the merged update when the buffer threshold is back to normal.
The market data traffic management algorithm in the ADS allows the ADS to slow down the stream of updates to a client application. Enabling this algorithm should significantly reduce the occurrence of channel disconnects due to buffer overflow at the cost of merging multiple update events for an instrument.
In both cases, Tick-by-Tick data will be lost.
