Relational OLTP: point reads and concurrent writes, on disk, no contention tax
You are looking at Bitcoin value-flow188.97M nodes as 64-byte containers in one Semurg substrate. Ask it to SELECT a tx/address by id, or INSERT rows concurrently and it runs the
real query live on this box, on data you can
download and re-run yourself.
Click an example in the agent rail, or type SELECT for point reads, or INSERT for concurrent writes. Every run is measured end to end, on
this box, on the real dataset. Nothing is pre-computed.
A row is a 64-byte container keyed by its tx/address id. A point SELECT is a lock-free seqlock read off the RAM-ring (no buffer-pool latch); an INSERT is a group-committed append to a lock-free write ring (no per-row WAL fsync). Both fan across every core. Total time, measured live.
the transactional row store
LATCH AND FSYNC CONTENTION. Point reads serialize on a buffer-pool latch and page pin; concurrent commits serialize on a single WAL fsync. Add cores and the throughput curve flattens.
semurg, one engine
Lock-free reads, group-commit writes. A SELECT is a seqlock load with no latch; an INSERT is a write-ring append with no per-row fsync. Both fan across every core and scale up, not flatten.
No buffer pool, no WAL, no read replica, no cache out front. One durable engine, fast like memory, on disk.
The Bitcoin value-flow table, 188.97M tx/address nodes, point SELECTs and concurrent INSERTs fanned across every core. Measured live on this box. Re-run it yourself.
ask the agent to SELECT (point reads) or INSERT (concurrent writes), the live throughput renders here
Honest caveat: reads are shown cold AND warm so this is never our-hot-versus-their-cold, and both numbers are total round-trip time with no asterisk. The point read is a lock-free seqlock load with no buffer-pool latch; the write is a group-committed append with no per-row WAL fsync, which is precisely the contention that flattens a transactional SQL engine under high concurrency. The additive nuance: this is one durable engine that also serves graph, full-text and vectors, so the collapse removes the machinery rather than tuning it.