Async Data Fetch vs Thread Pool Tuning: Which Better Controls Latency for Concurrent Requests in Jira Align?

Answer

Under peak concurrent request volume, tuning the Jira Align thread pool typically yields the greater latency reduction. Asynchronous data fetching only adds a noticeable benefit when a significant portion of request time is spent waiting on I/O (e.g., database or external service calls).

Confirmed Facts

• Thread pool size and queue length directly limit how many request threads can run simultaneously, which bounds latency when the workload is CPU‑bound.
• Async data fetch (non‑blocking I/O) reduces thread blocking only for the I/O‑wait portion of a request.
• In typical Jira Align workloads, CPU‑bound processing dominates, making thread pool tuning the primary lever for latency.

Likely Explanation (Conditional on Profiling)

If profiling shows that I/O wait exceeds roughly 30 % of total request time, enabling async servlet/completable‑future processing can further lower latency by freeing threads while waiting. Otherwise, the latency gain from async fetch is marginal compared to adjusting the thread pool.

Steps for This Case

1. Run a realistic load test that mimics peak concurrent users.
2. Measure request latency and thread‑pool metrics (active threads, queue length, rejected tasks).
3. Profile CPU vs. I/O wait using a Java profiler or OS tools (e.g., jstack, async‑profiler, or perf).
4. If I/O wait ≤ 30 %:
1. Increase the thread pool max threads (and optionally queue size) until latency stops improving or context‑switch overhead rises.
2. Re‑run the load test to verify latency change.
5. If I/O wait > 30 %:
1. First tune the thread pool as above to handle the CPU‑bound portion.
2. Enable asynchronous data fetching for DB/service calls (e.g., switch to CompletableFuture or async servlet).
3. Re‑test to confirm additional latency reduction.
6. Monitor for side effects: excessive thread‑pool size can increase context‑switching overhead; async APIs add complexity in error handling and transaction management.

Missing Diagnostic Detail

To decide whether async fetch should be enabled, we need the observed I/O wait percentage under load. Please provide the I/O wait proportion (or average blocking time) from your profiling data.