Index performance changes when application work affects cache and memory
Revisiting Index Performance under High Cache-Miss Penalties
Databases
Summary
Large data applications rely on indexes to quickly find information, but how well indexes work depends on what happens after the lookup. The authors show that testing indexes by themselves can be misleading because real applications use memory and CPU after looking up data. They created a new test called AuxBench to see how indexes perform alongside other work and found that some designs that do well alone struggle when used with more application activity. They also discovered that request patterns can cause unexpected delays near peak usage.
What this means in practice
- •For database system engineers: Evaluate index designs more realistically by including post-lookup memory activity to choose better indexes for large data applications.
- •For high-performance computing teams: Optimize workload scheduling by understanding how varying request arrival patterns affect index latency under load.
Authors
Mathew M. Kim, Beomseok Nam
Abstract
Modern data-intensive applications often query very large datasets, making index performance critical. Recent index structures exploit the fine-grained access capabilities of emerging storage devices such as DCPMM and CXL. However, existing benchmarks typically evaluate indexes in isolation using saturated, closed-loop workloads that emphasize maximum throughput. Such workloads favor highly concurrent designs while overlooking the application work that follows an index lookup and competes for cache, memory bandwidth, and CPU resources. We present AuxBench - a microbenchmark that evaluates indexes as auxiliary access paths within larger application workloads. Using AuxBench, we evaluate four byte-addressable indexes under controlled post-lookup memory activity, concurrency, and request arrival patterns. Our experiments show that post-lookup memory activity can change the relative performance of competing indexes, that concurrency-oriented designs may lose their advantages when substantial application work is introduced, and that time-varying request arrivals can produce severe tail latency near saturation. These findings demonstrate why index-only benchmarks may not adequately predict performance in larger application workloads