Refactoring the foundation of financial computing power to help quantitative giants gain millisecond level trading opportunities!!
A financial technology company in Shanghai was established in 2016, deeply engaged in the field of quantitative asset management. With strong data mining and algorithm development capabilities, it has built a cross market, multi variety intelligent trading platform and continues to lead the domestic high-frequency trading track. The company takes technological innovation as its core competitiveness and is committed to breaking through the limits of market microstructure games through computing power upgrades.
<0.1ms
End to end transaction latency compressed to within 0.1 milliseconds
Processing speed
Customized FPGA+10 Gigabit Ethernet card achieves an 8-fold increase in order processing efficiency
Strategic return
Alpha factor mining cycle shortened by 70%, strategy annualized returns increased by 22%
Project Background
A financial technology company in Shanghai was established in 2016, deeply engaged in the field of quantitative asset management. With strong data mining and algorithm development capabilities, it has built a cross market, multi variety intelligent trading platform and continues to lead the domestic high-frequency trading track. The company takes technological innovation as its core competitiveness and is committed to breaking through the limits of market microstructure games through computing power upgrades.
Challenge pain points
Fragmentation of computing power resources
Multi market parallel backtesting requires over 2000 concurrent computing cores, and the efficiency of existing infrastructure resource scheduling is less than 30%.
System stability risk
Hardware failures or software anomalies may trigger millions of daily losses, requiring strict requirements for system redundancy design and fault tolerance mechanisms.
The cost of algorithm iteration is high
Strategy optimization relies on TB level historical data training, and traditional architectures take over 48 hours for a single backtesting, severely limiting the speed of strategy iteration.
Solution
Heterogeneous computing power resource pool
Build a CPU+GPU+FPGA hybrid computing cluster that supports 2000 concurrent backtesting cores, improves resource utilization to 95%, and shortens strategy development cycle by 65%.
Multi disaster recovery architecture
Adopting dual active data center deployment and hardware level redundancy design, the system availability reaches 99.999%, and the fault switching time is less than 10ms.
Distributed memory database
Equipped with 8TB Optane persistent memory, it supports real-time loading of TB level market data and improves backtesting efficiency by 12 times.
Project Benefits
Order execution delay reduced by 80%, high-frequency strategy sliding point control accuracy improved to 0.01%, and annualized revenue increased; Heterogeneous computing power clusters support daily policy iteration, with a 300% increase in Alpha factor mining efficiency and an expansion of policy library capacity to over 500; The system operates without any malfunctions throughout the year, with a 100% self-healing rate for abnormal events and a 40% reduction in maintenance costs;
Customer reviews
The high-frequency server selected this time exceeded expectations, greatly improving the delay problem in the information transmission process. It not only optimized the transaction execution speed, but also won valuable opportunities in the market
