Why do multiple jobs reading the same NativeArray run serially?

Without [ReadOnly], scheduler assumes write access. Mark fields [ReadOnly]; multiple jobs can then read in parallel.

What if some jobs write?

Writers serialize against any reader/writer. Plan: do all reads in parallel batch, then a writer pass with handle dependency.

For NativeList growing across jobs, AsDeferredJobArray defers length resolution until job runs. Useful for producer-consumer patterns.

Fix: Multiple Jobs Reading Same Data Serialize

Quick answer: Mark [ReadOnly] on shared NativeArray fields in each job. Jobs run in parallel.

Two analysis jobs read the same dataset. Profiler shows them running serially. Without ReadOnly, scheduler enforces ordering.

The Fix

[BurstCompile]
struct JobA : IJob {
    [ReadOnly] public NativeArray<float> SharedData;
    public NativeArray<float> OutA;
    public void Execute() { /* read */ }
}

[BurstCompile]
struct JobB : IJob {
    [ReadOnly] public NativeArray<float> SharedData;   // same array
    public NativeArray<float> OutB;
    public void Execute() { /* read */ }
}

// Schedule both in parallel
var hA = jobA.Schedule();
var hB = jobB.Schedule();   // not dependent on hA
JobHandle.CompleteAll(ref hA, ref hB);

ReadOnly tells the safety system multiple readers are fine. Without it: serial.

Verifying

Profiler: jobA and jobB on different worker threads simultaneously. Without ReadOnly: sequential.

“ReadOnly. Parallel reads.”

Related Issues

For NativeList resize, see resize. For ReadOnly write error, see readonly write.

ReadOnly. Parallel reads.