Performance across execution and storage engines.

Results are shown with their workload, environment and method so they can be interpreted and reproduced.

5.40Mhot reads / sec88.6krandom updates / sec14.839 s20M cold replay8.00 MiBsmallest footprint

01 / EXECUTION

One BASIC program, three methods.

Logical operations per second. MVB bytecode and Java execution are Professional capabilities; the Java figure records the selected compiled strategy for this workload.

Arithmetic

Source455,639MVB bytecode471,087Java251,483

Branch-heavy

Source337,149MVB bytecode379,332Java164,420

Dynamic array

Source10,035MVB bytecode34,200Java27,781

String manipulation

Source360,063MVB bytecode706,569Java120,676

Successful warm reads / second

WorkloadReplayMapDBBlockUberHashLogStore

Sequential warm read2,736,4122,433,096189,343218,2263,748,238

Random warm read3,118,7922,173,653188,989268,0603,617,422

Hot-record warm read4,543,3892,870,396197,925440,7235,401,026

Writes and mutations / second

WorkloadReplayMapDBBlockUberHashLogStore

Initial load280,87824,62445,94849,44360,440

Random update97,92010,86327,02588,55538,267

Sequential delete323,50916,42622,9834,87088,728

02 / SCALE RUN

Twenty million records.

Buffered Replay loaded twenty million tiny readings. Cold startup reconstructed 19,990,000 live items from the append log.

Hot-record warm read10,484,972Sequential warm read1,747,752Random warm read1,640,799Initial load1,508,741Random update961,073Sequential delete553,879

Final disk footprint

MapDB8.00 MiBUberHash9.81 MiBReplay10.06 MiBBlock17.64 MiBLogStore47.97 MiB

METHOD / 21 JULY 2026

Host: Intel Core i9-13900K · 24 cores / 32 threads · 95.8 GiB RAM
Runtime: Windows 11 · OpenJDK 21.0.2
Standard set: 10,000 records · approximately 512 bytes each
Scope: Storage and logical execution workloads; not end-to-end IES transactions