SIMBA SPECTRA: Implementation Details and Architecture
Advantages of Simple Binary Encoding
SIMBA protocol is based on FIX Simple Binary Encoding to simplify codec development, especially for FPGA systems.
SBE represents data in order similar to its’ order in memory - this allows to directly analyze the value of a specific field in many cases. This feature is especially valuable for HFT and ultra-low latency trading systems as it provides an ability to react to market signals faster.
Moving towards binary encoding is in line with the best practices of leading exchanges. Furthermore, TWIME trading interface for SPECTRA trading platform also uses binary encoding, so combining these two services may simplify clients’ front-end systems development and provide an ability for deeper integration of its’ components.
New best prices in the very beginning of a transaction
Every new order that reaches the matching engine produces a "transaction" - a list of order book modifications caused by that new order. A transaction is atomic and published only after all the order book modifications has been done.
FAST FOL gate consecutively publishes order book events, which must be processed one by one to find the new final order book state or estimate the price shift.
SIMBA publishes new best bid&ask prices in the very beginning of a transaction. This gives traders equal opportunities to estimate the magnitude of price movement.
Innovations in software part
The main difference between SIMBA SPECTRA and other services (FAST FOL, TWIME) architecture is a direct connection of SIMBA Gate to the Matching core, bypassing the Persistence layer. This reduces the overall latency by ≈10 microseconds (preliminary internal measurements).
The new 10G network infrastructure is implemented in MOEX Co-location for low latency market data services (currently only SIMBA for Derivatives market, but in future the similar services may be launched for Equities&Bonds and FX markets). This segment serves market data traffic exclusively eliminating interference with trading and auxiliary services’ traffic. The network is built on industry-standard equipment supplied by a well-known manufacturer.
One-way latency from client-facing access switch to server-facing access switch is ≈10 µs lower than for FAST FOL in standard network (preliminary internal measurements).
The new dedicated 10G link to low-latency market data network in MOEX Co-location is required to use the service. Both L2 and L3 connections are available.