Loading...

XML

Word

Printable

JSON

Performance benchmark phase T8b (~5 days). Throughput and flush overhead benchmarking.

Test coverage:

Benchmark bulk insert flush throughput at 1M, 10M, 100M rows
Compare against: regular cursor insert, existing WiredTiger bulk loader, bulk insert with flush disabled
Measure flush overhead: extra latency per flush, write amplification (bytes flushed / bytes inserted)
Measure checkpoint impact: does periodic flush reduce checkpoint duration vs one giant flush at end?
Report: rows/sec, MB/sec, flush count, flush latency p50/p99, peak RSS

Depends on: WT-17215 (T4)
Can run in parallel with: T8a (WT-17221)

is related to

WT-17215 Bulk insert flush — implement production __wt_bulk_flush() orchestrator

WT-17218 Bulk insert flush — cppsuite stress test: concurrent bulk loads with background checkpoint

WT-17219 Bulk insert flush — cppsuite stress test: crash injection during flush

WT-17220 Bulk insert flush — cppsuite stress test: memory pressure

related to

WT-17221 Bulk insert flush — scale test: large table load and O(1) memory verification