The cost of throwing exceptions in C#

This repository contains a small project showing the cost of throwing exceptions in C#. It examines a very common case: An entity (in this case, a user) is requested from a service, but the service cannot find the entity. How should the service communicate this?

The project includes three methods that each communicates the result in a different way.

1. GetUserWithExceptions(): A method that will throw an exception if the user is not found.
2. GetUserWithDefault(): A method that will return default if the user is not found.
3. GetUserWithTryGet(): A method that will return true or false to indicate if the user was found or not, and - if the user was found - return him in an out variable.

Benchmarking results

BenchmarkDotNet is used to benchmark the three methods.

Using each of the three methods, the benchmark attempts to fetch a non-existent user from the user service. For the sake of curiosity, I have used .NET 4.8, .NET Core 2.1., .NET Core 3.1 and .NET 5.0 as runtimes.

The results of the benchmark can be seen in the table below.

BenchmarkDotNet=v0.12.1, OS=Windows 10.0.19042
Intel Core i7-8700K CPU 3.70GHz (Coffee Lake), 1 CPU, 12 logical and 6 physical cores
  [Host]     : .NET Framework 4.8 (4.8.4300.0), X64 RyuJIT
  Job-ICSTSG : .NET Framework 4.8 (4.8.4300.0), X64 RyuJIT
  Job-OYULFM : .NET Core 2.1.23 (CoreCLR 4.6.29321.03, CoreFX 4.6.29321.01), X64 RyuJIT
  Job-ZTBEOE : .NET Core 3.1.10 (CoreCLR 4.700.20.51601, CoreFX 4.700.20.51901), X64 RyuJIT
  Job-NTTAIL : .NET Core 5.0.1 (CoreCLR 5.0.120.57516, CoreFX 5.0.120.57516), X64 RyuJIT

IterationCount=30  LaunchCount=3  WarmupCount=10

|                Method |       Runtime |          Mean |       Error |      StdDev |        Median |    Ratio | RatioSD |
|---------------------- |-------------- |--------------:|------------:|------------:|--------------:|---------:|--------:|
|    GetUserWithDefault |      .NET 4.8 |      8.707 ns |   0.0959 ns |   0.2672 ns |      8.590 ns |     1.00 |    0.00 |
|     GetUserWithTryGet |      .NET 4.8 |      9.197 ns |   0.1616 ns |   0.4424 ns |      9.258 ns |     1.06 |    0.06 |
| GetUserWithExceptions |      .NET 4.8 | 14,500.979 ns |  53.0287 ns | 143.3661 ns | 14,506.671 ns | 1,664.59 |   48.59 |
|                       |               |               |             |             |               |          |         |
|     GetUserWithTryGet | .NET Core 2.1 |      6.878 ns |   0.0292 ns |   0.0789 ns |      6.899 ns |     0.92 |    0.04 |
|    GetUserWithDefault | .NET Core 2.1 |      7.499 ns |   0.1088 ns |   0.2961 ns |      7.339 ns |     1.00 |    0.00 |
| GetUserWithExceptions | .NET Core 2.1 | 14,962.689 ns | 154.2974 ns | 427.5574 ns | 14,913.022 ns | 1,997.56 |  114.99 |
|                       |               |               |             |             |               |          |         |
|    GetUserWithDefault | .NET Core 3.1 |      6.965 ns |   0.0212 ns |   0.0567 ns |      6.936 ns |     1.00 |    0.00 |
|     GetUserWithTryGet | .NET Core 3.1 |      7.814 ns |   0.0195 ns |   0.0504 ns |      7.797 ns |     1.12 |    0.01 |
| GetUserWithExceptions | .NET Core 3.1 | 13,392.766 ns |  63.8685 ns | 172.6721 ns | 13,303.902 ns | 1,923.56 |   32.22 |
|                       |               |               |             |             |               |          |         |
|     GetUserWithTryGet | .NET Core 5.0 |      6.320 ns |   0.0382 ns |   0.1051 ns |      6.334 ns |     0.95 |    0.05 |
|    GetUserWithDefault | .NET Core 5.0 |      6.682 ns |   0.1305 ns |   0.3573 ns |      6.517 ns |     1.00 |    0.00 |
| GetUserWithExceptions | .NET Core 5.0 | 13,241.254 ns |  76.2873 ns | 211.3918 ns | 13,232.031 ns | 1,985.94 |  104.29 |

(Take note of the time scale of nanoseconds when comparing the results.)

The table shows the results of benchmarking the three methods in each of the previously mentioned runtimes. Ratio is the most interesting column and shows the performance of the specified method as a ratio of the baseline method performance (in this case, the baseline is GetUserWithDefault()).

The results show that the differences between GetUserWithDefault() and GetUserWithTryGet() are tiny. The differences should probably be attributed to uncertainties in measurement rather than an actual performance variations.

It is however obvious that there is a great performance penalty for throwing exceptions: GetUserWithExceptions() is ~1700 to ~2000 times slower than its counterparts.