WorkflowForge Competitive Benchmark Analysis

Version: 2.1.1
Analysis Date: March 2026
Frameworks Tested:

  • WorkflowForge 2.1.1
  • Workflow Core
  • Elsa Workflows

Test System: Windows 11 (25H2), Intel 11th Gen i7-1185G7, .NET SDK 10.0.103
Runtimes: .NET 10.0.3, .NET 8.0.24, .NET Framework 4.8.1
BenchmarkDotNet: v0.15.8 (50 iterations, 5 warmup)
Benchmark Run: March 7, 2026

Table of Contents

Executive Summary

Across twelve benchmark scenarios on .NET 10.0, .NET 8.0, and .NET Framework 4.8 (50 iterations each), WorkflowForge measured 13–511x faster execution and 6–575x less allocation than Workflow Core and Elsa for the same scripted logic.

Metric Value
Max Speed Advantage 511x faster (State Machine 25 transitions, .NET 10.0)
Max Memory Advantage 575x less allocation (Parallel 16 ops, .NET 10.0)
Min Execution Time 11μs (Creation Overhead, .NET 10.0)
Min Memory Footprint 3.6KB
511x
Faster (State Machine)
575x
Less Memory
11μs
Min Execution Time
3.6KB
Min Memory

Recorded ranges (same hardware, shared scripts):

  • WorkflowForge medians sit in the microsecond band (11–706μs) on these runs; Workflow Core and Elsa land in milliseconds (0.3–109ms) for the same scenarios.
  • Reported WorkflowForge allocations stay in kilobytes (3.5–256KB) while competitors often allocate megabytes (0.04–19MB) in the same tests.
  • The State Machine sweep posts the largest execution gap: up to ~511x versus Elsa on .NET 10.0 for 25 transitions.
  • Concurrent Execution spans 118–288x faster than Elsa across the three runtimes we tested.
  • Sequential scenarios measure 26–55x faster with comparatively small memory use.
  • The pattern holds across all twelve scenario types and three runtimes (Elsa omitted on .NET Framework 4.8 where unsupported).

Visual Performance Comparison

Execution Time (Lower is Better)

Runtime Scenario WorkflowForge Workflow Core Elsa WF Advantage
.NET 10.0 State Machine (25) 65μs 29,537μs 33,062μs 455-511x
.NET 8.0 State Machine (25) 71μs 21,683μs 34,426μs 305-485x
.NET FX 4.8 State Machine (25) 61μs 18,486μs N/A† 303x
.NET 10.0 Concurrent (8 wf) 372μs 47,114μs 87,491μs 127-235x
.NET 8.0 Concurrent (8 wf) 357μs 42,054μs 103,024μs 118-288x
.NET FX 4.8 Concurrent (8 wf) 167μs 41,934μs N/A† 250x
.NET 10.0 Sequential (10 ops) 422μs 13,828μs 18,676μs 33-44x
.NET 8.0 Sequential (10 ops) 377μs 9,879μs 19,168μs 26-51x
.NET FX 4.8 Sequential (10 ops) 122μs 6,743μs N/A† 55x
State Machine Execution (25 Transitions)
State machine scenario, .NET 10.0 (see table)
65μs
29.5ms
33.1ms
.NET 10.0
71μs
21.7ms
34.4ms
.NET 8.0
61μs
18.5ms
.NET FX 4.8
WorkflowForge
Workflow Core
Elsa Workflows

Memory Allocation (Lower is Better)

Runtime Scenario WorkflowForge Workflow Core Elsa WF Advantage
.NET 10.0 Concurrent (8 wf) 155 KB 3,247 KB 19,568 KB 21-126x
.NET 8.0 Concurrent (8 wf) 155 KB 3,308 KB 19,572 KB 21-126x
.NET FX 4.8 Concurrent (8 wf) 272 KB 3,816 KB N/A† 14x
.NET 10.0 Parallel (16 ops) 8.0 KB 126 KB 4,576 KB 16-575x
.NET 8.0 Parallel (16 ops) 8.2 KB 125 KB 4,651 KB 15-567x
Memory Allocation - Concurrent Execution (8 Workflows)
Concurrent run: WorkflowForge ~155KB reported vs ~3.2MB / ~19.6MB in this harness
155KB
3.2MB
19.6MB
.NET 10.0
155KB
3.2MB
19.6MB
.NET 8.0
272KB
3.8MB
.NET FX 4.8
WorkflowForge
Workflow Core
Elsa Workflows

Scaling charts (workload sweeps)

Measured WorkflowForge gap widens as the scripted workload grows in these sweeps.

Scenario Scale WF vs Elsa
Sequential 1 op → 50 ops 35.8x → 95.4x
Loop/ForEach 10 items → 100 items 71.8x → 156.0x
Concurrent 1 wf → 8 wf 74.2x → 288.3x
Conditional 10 ops → 50 ops 64.3x → 109.6x
Execution Scaling - Advantage Grows with Workload
Larger scripted workloads; ratios from the sweep tables
257μs
1.9ms
9.2ms
Seq 1 op
615μs
31.8ms
58.6ms
Seq 50 ops
270μs
7.2ms
19.4ms
Loop 10
660μs
60.2ms
103ms
Loop 100
260μs
7.6ms
19.3ms
Conc 1 wf
357μs
42.1ms
103ms
Conc 8 wf
301μs
9.2ms
19.4ms
Cond 10
526μs
32.1ms
57.7ms
Cond 50
WorkflowForge
Workflow Core
Elsa Workflows

Scenario Breakdown

Scenario 1: Simple Sequential Workflow

Description: Execute operations sequentially (1, 5, 10, 25, 50 operations)

Multi-Runtime Performance (Median, 10 ops)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 422μs 13,828μs 18,676μs
.NET 8.0 377μs 9,879μs 19,168μs
.NET FX 4.8 122μs 6,743μs N/A†

Parameter Sweep (.NET 8.0)

Operations WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
1 257μs 1,878μs 9,175μs 7.3x faster 35.8x faster
5 319μs 6,078μs 14,168μs 19.0x faster 44.4x faster
10 377μs 9,879μs 19,168μs 26.2x faster 50.9x faster
25 462μs 27,075μs 34,395μs 58.6x faster 74.5x faster
50 615μs 31,768μs 58,648μs 51.7x faster 95.4x faster

Memory Allocation (10 ops, by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 17.72KB 427KB 3,024KB
.NET 8.0 17.72KB 429KB 2,992KB
.NET FX 4.8 40.00KB 560KB N/A†

Memory Allocation - Parameter Sweep (.NET 8.0)

Operations WorkflowForge Workflow Core Elsa
1 3.98KB 46KB 1,254KB
5 10.07KB 218KB 2,018KB
10 17.72KB 429KB 2,992KB
25 48.93KB 1,064KB 5,956KB
50 83.86KB 2,126KB 10,879KB

Numbers: the execution gap versus Workflow Core and Elsa widens as the operation count rises in the sequential sweep.

Scenario 2: Data Passing Workflow

Description: Pass data between operations (5, 10, 25 operations)

Multi-Runtime Performance (Median, 10 ops)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 325μs 11,651μs 18,510μs
.NET 8.0 321μs 9,751μs 19,164μs
.NET FX 4.8 118μs 6,684μs N/A†

Parameter Sweep (.NET 8.0)

Operations WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
5 299μs 5,063μs 14,052μs 16.9x faster 47.0x faster
10 321μs 9,751μs 19,164μs 30.4x faster 59.7x faster
25 483μs 17,318μs 33,825μs 35.9x faster 70.0x faster

Memory Allocation (10 ops, by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 16.36KB 425KB 3,024KB
.NET 8.0 16.36KB 429KB 2,988KB
.NET FX 4.8 40.00KB 544KB N/A†

Memory Allocation - Parameter Sweep (.NET 8.0)

Operations WorkflowForge Workflow Core Elsa
5 9.45KB 216KB 2,018KB
10 16.36KB 429KB 2,988KB
25 39.26KB 1,063KB 5,956KB

What matters: extra data-passing steps add under about a microsecond per operation in WorkflowForge for this scenario.

Scenario 3: Conditional Branching

Description: Conditional logic with if/else branches (10, 25, 50 operations)

Multi-Runtime Performance (Median, 10 ops)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 333μs 13,427μs 19,166μs
.NET 8.0 301μs 9,248μs 19,361μs
.NET FX 4.8 118μs 6,562μs N/A†

Parameter Sweep (.NET 8.0)

Operations WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
10 301μs 9,248μs 19,361μs 30.7x faster 64.3x faster
25 382μs 16,844μs 33,480μs 44.1x faster 87.6x faster
50 526μs 32,140μs 57,654μs 61.1x faster 109.6x faster

Memory Allocation (10 ops, by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 19.48KB 424KB 3,019KB
.NET 8.0 19.48KB 427KB 2,991KB
.NET FX 4.8 48.00KB 552KB N/A†

Memory Allocation - Parameter Sweep (.NET 8.0)

Operations WorkflowForge Workflow Core Elsa
10 19.48KB 427KB 2,991KB
25 48.04KB 1,061KB 5,947KB
50 88.97KB 2,121KB 10,907KB

Takeaway: branch decisions add negligible time in WorkflowForge here (under ~1μs per decision in the harness).

Scenario 4: Loop/ForEach Processing

Description: Iterate over collections (10, 50, 100 items)

Multi-Runtime Performance (Median, 50 items)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 450μs 35,320μs 54,827μs
.NET 8.0 495μs 30,742μs 58,347μs
.NET FX 4.8 350μs 34,137μs N/A†

Parameter Sweep (.NET 8.0)

Items WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
10 270μs 7,242μs 19,404μs 26.8x faster 71.8x faster
50 495μs 30,742μs 58,347μs 62.1x faster 117.9x faster
100 660μs 60,218μs 102,879μs 91.2x faster 156.0x faster

Memory Allocation (50 items, by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 96.93KB 2,086KB 10,907KB
.NET 8.0 96.34KB 2,121KB 10,907KB
.NET FX 4.8 176.00KB 2,512KB N/A†

Memory Allocation - Parameter Sweep (.NET 8.0)

Items WorkflowForge Workflow Core Elsa
10 20.48KB 428KB 2,985KB
50 96.34KB 2,121KB 10,907KB
100 194.85KB 4,241KB 20,859KB

Numbers: the ForEach scenario shows a larger speedup as the collection grows from 10 to 100 items.

Scenario 5: Concurrent Execution

Description: Execute multiple workflows concurrently (1, 4, 8 workflows)

Multi-Runtime Performance (Median, 8 workflows)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 372μs 47,114μs 87,491μs
.NET 8.0 357μs 42,054μs 103,024μs
.NET FX 4.8 167μs 41,934μs N/A†

Parameter Sweep (.NET 8.0)

Concurrency WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
1 260μs 7,588μs 19,265μs 29.2x faster 74.2x faster
4 322μs 21,717μs 56,360μs 67.4x faster 175.0x faster
8 357μs 42,054μs 103,024μs 117.8x faster 288.3x faster

Memory Allocation (8 workflows, by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 154.66KB 3,247KB 19,568KB
.NET 8.0 154.67KB 3,308KB 19,572KB
.NET FX 4.8 272.00KB 3,816KB N/A†

Memory Allocation - Parameter Sweep (.NET 8.0)

Concurrency WorkflowForge Workflow Core Elsa
1 20.00KB 426KB 2,983KB
4 79.68KB 1,627KB 9,861KB
8 154.67KB 3,308KB 19,572KB

What matters: total time scales with the number of workflows, but per-workflow overhead stays flat in WorkflowForge as concurrency rises in this test.

Scenario 6: Error Handling

Description: Exception handling and recovery

Multi-Runtime Performance (Median)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 70μs 1,498μs 7,694μs
.NET 8.0 114μs 1,349μs 7,737μs
.NET FX 4.8 88μs 4,471μs N/A†

Memory Allocation (by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 7.02KB 51KB 1,056KB
.NET 8.0 8.38KB 47KB 1,072KB
.NET FX 4.8 N/A‡ 864KB N/A†

Advantage: 13-110x faster than competitors, 6-150x less memory.

Takeaway: error-handling paths finish in about 70–114μs median on WorkflowForge versus milliseconds on the others in this scenario.

Scenario 7: Creation Overhead

Description: Workflow instantiation cost

Multi-Runtime Performance (Median)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 11μs 1,001μs 2,245μs
.NET 8.0 11μs 819μs 2,328μs
.NET FX 4.8 7μs 260μs N/A†

Memory Allocation (by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 3.72KB 125KB 537KB
.NET 8.0 3.72KB 129KB 578KB
.NET FX 4.8 N/A‡ 128KB N/A†

Advantage: 37-206x faster than competitors, 33-155x less memory.

Numbers: creating a workflow instance costs about 7–11μs median on WorkflowForge in the creation-overhead benchmark.

Scenario 8: Complete Lifecycle

Description: Full create-execute-dispose cycle (Workflow Core excluded)

Multi-Runtime Performance (Median)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 36μs N/A 9,877μs
.NET 8.0 59μs N/A 9,723μs
.NET FX 4.8 33μs N/A N/A†

Memory Allocation (by Runtime)

Runtime WorkflowForge Elsa
.NET 10.0 3.69KB 1,513KB
.NET 8.0 3.69KB 1,510KB
.NET FX 4.8 N/A‡ N/A†

Advantage: 165-274x faster than Elsa, 410x less memory.

Note: Workflow Core was excluded from this benchmark because WorkflowHost.Start() starts background worker threads meant to run continuously, which does not line up with tight create-start-stop-dispose loops at 50 iterations. That is a model mismatch for this particular test, not a claim about raw CPU speed.

What matters: WorkflowForge completes the full create-execute-dispose cycle in about 33–59μs median here.

Scenario 9: State Machine

Description: State machine with multiple transitions (5, 10, 25 transitions)

Multi-Runtime Performance (Median, 25 transitions)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 65μs 29,537μs 33,062μs
.NET 8.0 71μs 21,683μs 34,426μs
.NET FX 4.8 61μs 18,486μs N/A†

Parameter Sweep (.NET 8.0)

Transitions WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
5 36μs 6,275μs 14,444μs 174.3x faster 401.2x faster
10 43μs 10,028μs 19,626μs 233.2x faster 456.4x faster
25 71μs 21,683μs 34,426μs 305.4x faster 484.9x faster

Memory Allocation (25 transitions, by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 23.92KB 1,090KB 5,966KB
.NET 8.0 23.92KB 1,105KB 5,937KB
.NET FX 4.8 24.00KB 1,344KB N/A†

Memory Allocation - Parameter Sweep (.NET 8.0)

Transitions WorkflowForge Workflow Core Elsa
5 5.45KB 261KB 2,017KB
10 8.65KB 472KB 2,986KB
25 23.92KB 1,105KB 5,937KB

Takeaway: the state-machine sweep posts the largest execution deltas in the suite (up to ~511x versus Elsa on .NET 10.0 for 25 transitions).

Scenario 10: Long Running

Description: Long-running operations with delays (delay-bound scenario)

Multi-Runtime Performance (Median, 5 ops, 5ms delay)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 71,678μs 70,868μs 83,599μs
.NET 8.0 71,885μs 70,672μs 82,982μs
.NET FX 4.8 76,447μs 75,129μs N/A†

Parameter Sweep (.NET 8.0)

Ops/Delay WorkflowForge Workflow Core Elsa
3 ops/1ms 38,836μs 38,599μs 50,641μs
5 ops/1ms 71,802μs 70,252μs 81,715μs
5 ops/5ms 71,885μs 70,672μs 82,982μs

Memory Allocation (by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 5.12KB 267KB 2,246KB
.NET 8.0 5.12KB 266KB 2,216KB
.NET FX 4.8 N/A‡ 393KB N/A†

Advantage: Similar timing (delay-bound); advantage is in 52-439x less memory.

Numbers: long-running scenarios are delay-bound, so wall-clock time tracks the configured delays. The measurable gap is allocation, not raw orchestration time.

Scenario 11: Parallel Execution

Description: Parallel operation execution within a workflow (4, 8, 16 operations)

Multi-Runtime Performance (Median, 16 ops, 4 concurrency)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 56μs 2,861μs 24,638μs
.NET 8.0 63μs 2,654μs 24,940μs
.NET FX 4.8 35μs 1,754μs N/A†

Parameter Sweep (.NET 8.0)

Ops/Concurrency WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
4 ops/2 68μs 2,771μs 13,147μs 40.8x faster 193.3x faster
8 ops/4 72μs 2,736μs 13,546μs 38.0x faster 188.1x faster
16 ops/4 63μs 2,654μs 24,940μs 42.1x faster 395.9x faster

Memory Allocation (by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 7.96KB 126KB 4,576KB
.NET 8.0 8.23KB 125KB 4,651KB
.NET FX 4.8 N/A‡ 184KB N/A†

What matters: parallel execution keeps a 38–396x execution advantage in this sweep alongside 15–575x less allocated memory versus the listed competitors.

Scenario 12: Event-Driven

Description: Event-driven workflow execution with delays

Multi-Runtime Performance (Median, 1ms delay)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 7,083μs 8,268μs 20,590μs
.NET 8.0 7,128μs 7,362μs 19,916μs
.NET FX 4.8 12,585μs 12,400μs N/A†

Parameter Sweep (.NET 8.0)

Delay WorkflowForge Workflow Core Elsa WF vs WC WF vs Elsa
1ms 7,128μs 7,362μs 19,916μs 1.0x 2.8x faster
5ms 7,147μs 8,613μs 20,608μs 1.2x faster 2.9x faster

Memory Allocation (by Runtime)

Runtime WorkflowForge Workflow Core Elsa
.NET 10.0 3.48KB 40KB 999KB
.NET 8.0 3.48KB 37KB 1,032KB
.NET FX 4.8 N/A‡ 90KB N/A†

Advantage: WorkflowForge and Workflow Core are near-parity on execution time (1ms delay); WorkflowForge is 2.8-2.9x faster vs Elsa. Memory advantage: 11-297x less.

Takeaway: event-driven runs are I/O-bound at the tested delays; WorkflowForge mainly separates from Elsa on memory use and steadier timing at 1–5ms delays.

Performance Advantage Summary

By Scenario Type (12 Scenarios)

# Scenario Speed Advantage Memory Advantage
1 Sequential (10 ops) 26-55x 24-171x
2 Data Passing (10 ops) 30-60x 26-185x
3 Conditional (10 ops) 31-64x 22-155x
4 Loop/ForEach (50 items) 62-118x 22-113x
5 Concurrent (8 workflows) 118-288x 21-126x
6 Error Handling 13-110x 6-150x
7 Creation Overhead 37-206x 33-155x
8 Complete Lifecycle 165-274x 410x
9 State Machine (25 trans) 303-511x 46-249x
10 Long Running ~1x (delay-bound) 52-439x
11 Parallel (16 ops) 38-396x 15-575x
12 Event-Driven 1.0-2.9x 11-297x

Ranges include all three runtimes (.NET 10.0, .NET 8.0, .NET Framework 4.8). Elsa is excluded from .NET Framework 4.8 comparisons.

Overall Speed Range: 13-511x faster execution (compute-bound scenarios)
Overall Memory Range: 6-575x less memory allocation

Reading the summary table

  1. State machine carries the widest execution spread we recorded (303–511x in the sweep).
  2. Concurrent work stays in the 118–288x band vs Elsa on the runtimes listed.
  3. Long running and event-driven rows are delay-heavy; the standout delta there is allocation (WorkflowForge 52–439x lower in those tests).
  4. On .NET 10.0 and 8.0, WorkflowForge reported less allocated memory in every row we logged (Elsa omitted on .NET Framework 4.8).

Architectural Differences

WorkflowForge Design

  1. Lightweight Execution Model
    • No background threads
    • Synchronous sequential execution (default)
    • Explicit parallelism via ForEachWorkflowOperation
    • Minimal object allocations
  2. Dictionary-Based Data Flow
    • ConcurrentDictionary<string, object?> for properties
    • Zero serialization overhead
    • Thread-safe property access
  3. Dependency-Free Core
    • No reflection-heavy frameworks
    • No serialization frameworks
    • Pure .NET Standard 2.0
  4. Middleware Pipeline
    • Russian Doll pattern
    • Minimal delegate allocations
    • No reflection per operation

Workflow Core Design

  1. Persistent Workflow Engine
    • Background worker threads
    • Designed for long-running workflows
    • Persistent state management
    • Work queue architecture
  2. Strong Typing
    • Reflection-based step resolution
    • JSON serialization for state
  3. Intended Use Case
    • Long-running business processes (hours/days)
    • Workflows that survive process restarts
    • Background processing

In short: Workflow Core targets durable, host-backed processes. WorkflowForge targets in-process, low-overhead runs first; add persistence packages when you need resume semantics.

Elsa Workflows Design

  1. Workflow Designer Focus
    • Visual workflow designer
    • HTTP workflow triggers
    • Large built-in activity catalog
  2. Serialization-Heavy
    • JSON serialization for all data
    • Heavy use of reflection
    • Large object graphs
  3. Intended Use Case
    • Visual workflow design
    • Human task workflows
    • Integration workflows

In short: Elsa emphasizes designer-first authoring, HTTP triggers, and a large built-in activity set. WorkflowForge keeps everything in C# with a smaller default surface. Neither layout is universally better; they optimize for different entry points.

Benchmark Methodology

Test Configuration

  • BenchmarkDotNet: v0.15.8
  • Runtimes: .NET 10.0.3, .NET 8.0.24, .NET Framework 4.8.1
  • Iterations: 50 per benchmark
  • Warmup: 5 iterations
  • Invocation: 1 per iteration
  • Unroll Factor: 1

Hardware

  • OS: Windows 11 (25H2)
  • CPU: Intel 11th Gen i7-1185G7
  • SDK: .NET SDK 10.0.103
  • Memory: Sufficient for all benchmarks

Scenario Implementations

All scenarios implement identical logic across all frameworks:

  • Same operation count
  • Same data structures
  • Same conditional logic
  • Same collection sizes
  • Same concurrency levels

Fairness Verification:

  • Workflow Core implementations use TaskCompletionSource for precise completion detection
  • Elsa implementations use proper workflow completion await
  • WorkflowForge implementations use standard ForgeAsync()

Reproduction

Full benchmark source code available in repository:

  • src/benchmarks/WorkflowForge.Benchmarks.Comparative/
  • All scenarios in Scenarios/ folder
  • Run via dotnet run -c Release

Statistical Significance

All results meet statistical significance criteria:

  • Standard deviation < 20% of mean (most scenarios)
  • P95 values show consistency
  • Median values used for comparison (more stable than mean)
  • 50 iterations provide statistical confidence

Outliers: Some scenarios show high standard deviation due to GC pauses or system activity. Median values are used to minimize impact.

Summary

Across these twelve scenarios, the harness logged 13–511x faster execution and 6–575x lower allocation for WorkflowForge vs Workflow Core and Elsa on .NET 10.0, 8.0, and .NET Framework 4.8. The deltas line up with a few concrete differences:

  1. Fewer moving parts in the default path: no host-owned worker pool, no baked-in durable store, no JSON round-trip on every hop.
  2. ConcurrentDictionary state instead of large per-step object graphs in this test code.
  3. Straight-line middleware: nested delegates instead of reflection-heavy resolution on each call.
  4. Different goals: WorkflowForge chases in-process throughput first; the other stacks bet on hosting models, designers, or long-running durability you may still want elsewhere.

On this hardware and these scripts, WorkflowForge had the lowest median times and allocations for programmatic, in-memory orchestration. Your workload and hosting still matter more than any ratio in a table.

References

  • WorkflowForge: https://github.com/animatlabs/workflow-forge
  • Workflow Core: https://github.com/danielgerlag/workflow-core
  • Elsa Workflows: https://github.com/elsa-workflows/elsa-core

† Elsa does not support .NET Framework 4.8; results are excluded for that runtime.
‡ BenchmarkDotNet does not report memory allocation metrics for .NET Framework 4.8 in some benchmark configurations.