Invalid input. Special characters are not supported.
In the fast-paced world of Financial Technology (FinTech), performance and efficiency are paramount. Stock exchange trading systems are a fundamental pillar of FinTech, designed to process massive volumes of financial data with exceptional speed and accuracy. In an environment where milliseconds can determine profitability, faster and lower-latency systems provide a decisive edge over competitors by enabling quicker order execution, reducing slippage, and capturing market opportunities ahead of others. This blog peeks into the findings from our analysis comparing Micron's MRDIMM (Multiplexed Rank DIMM) with traditional RDIMM (Registered Dual Inline Memory Module) in the context of an enterprise tick analysis workload, that is representative of a Fintech benchmark. The financial sector relies heavily on high-performance computing to manage and analyze vast amounts of data. Memory bandwidth and latency are critical factors that influence the performance of these systems.
Micron's MRDIMM@8800 MT/s delivers higher memory bandwidth and lower latencies at higher bandwidth compared to RDIMM@6400 MT/s, making it a compelling choice for memory-bound workloads.
STAC-M3 Shasta Benchmark
The Strategic Technology Analysis Center (STAC) Benchmark Council, comprising over 550 financial institutions and more than 60 vendor organizations, has developed the STAC-M3 benchmark suite. STAC-M3 is a set of industry-standard enterprise tick-analytics benchmarks and is used to evaluate database solutions that manage large time-series of market data (tick data). The STAC-M3 Shasta suite is a set of 15 benchmarks which focuses on memory performance by pre-loading data into memory and running various queries to measure access speed. Micron is collaborating with STAC on the analysis of MRDIMM and the benefits for FinTech workloads with a report underway.
Key Findings
Micron analyzed two specific enterprise tick-analytics benchmarks that are heavily memory bound (as measured by the Top-Down performance metrics collected with EMON which is part of the Intel vTune profiler) and gain significant improvement in memory bandwidth utilization and latencies with MRDIMM. These results have not been audited by STAC.
(Click image to enlarge)
Improvement with MRDIMM over RDIMM for the Highest Bid* benchmark
- Performance Improvement: The analysis revealed that MRDIMM@8800 MT/s offers up to 13-15% execution time speedup compared to RDIMM@6400 MT/s. This improvement is crucial for time-sensitive financial operations where every millisecond counts.
- Energy Efficiency: MRDIMM demonstrated up to 26% better task energy efficiency. As MRDIMM improves the execution time, the task-energy efficiency as measured by the amount of work done per watt over the runtime is significantly higher with MRDIMM, making it a more sustainable choice for data centers.
- Memory Bandwidth: MRDIMM provides 13% higher memory bandwidth utilization, which is essential for handling large datasets and complex queries efficiently.
- Latency: 25% better latency with MRDIMM. This is very helpful, especially in the context of the Financial Services Industry (FSI) where time is money.
The benchmarks included various queries against KDB+, an in-memory database, which showed significant improvements with MRDIMM. For example, the “Highest Bid”* benchmark, designed to test the efficiency of querying large datasets to find the highest bid price within a year with the added complexity of caching mechanisms, showed a 12% execution time speedup, 13% higher memory read bandwidth utilization and 25% lower memory latency with MRDIMM.
(Click image to enlarge)
Improvement with MRDIMM over RDIMM for the Statistics* benchmark
Similarly, the “Statistics”* benchmark which computes interval-based statistics, demonstrates a 15% execution time speedup, 8% higher memory read bandwidth utilization, and 27% lower memory latency.
As MRDIMM evolves, Micron Gen2 MRDIMM comes with higher memory speeds (@ 12800 MT/s) and better power efficiency, it continues to revolutionize the FinTech world with higher gains.
Conclusion
The analysis highlights the potential benefits of MRDIMM in enhancing the performance and efficiency of FinTech workloads. With its higher memory bandwidth, lower latency, and better energy efficiency, MRDIMM stands out as a superior choice for memory-bound applications in the financial sector. As the industry continues to evolve, adopting advanced memory solutions like MRDIMM will be crucial for staying ahead in the competitive FinTech landscape.
*The benchmarks are anonymized.
