dotnet/BenchmarkDotNet

TeBeCo

@tebeco

so i won't gamble
it's just what went through my mind
i might be wrong
just trying to isolate the code you want to bench

TeBeCo

@tebeco

@AndreyAkinshin
i got a very weird question ;)

is there a way, as of today to agregate bench results from differents project output

as of today (aspnetcore/signalr 3.1.2) has an hard dependency on MessagePack 1.7.x
and azure-signalr moved out by "building from source" using a submodule

we had an internal HubProtocol at work that relies on MessagePack too
I went for an update from MessagePack 1.x to 2.x (since we can now)

the thing is that i can't load twice the "same" Assembly with 2 different version to compare each other :

original MsgPack (from dotnet/aspnetcore) <= require 1.7.x
custom made <= require 1.x
custom made upgraded to 2.x

TeBeCo

@tebeco

also i'm curious about the way to change Name of a Parameterized bench ;)

the result look like this :

Adam Sitnik

@adamsitnik

also i'm curious about the way to change Name of a Parameterized bench

You need to override .ToString() in the class that is used as paramter (in your case its JsonHubProtocol etc)

TeBeCo

@tebeco

so i have to inherits the one from aspnetcore or wrap it ?

Adam Sitnik

@adamsitnik

you could just wrap it, or inherit. You can also add a new enum, use it as parameter and then handle in the setup method

sth like

public enum Hub
{
    Json,
    NetonsoftJson,
    MessagePack
}

[Params(Hub.Json, Hub.Json, Hub.MessagePack)]
public Hub Hub;

[GlobalSetup]
public Setup()
{
    switch (Hub)
    {
        case Json;
            hubProtocol = new JsonHubProtocol();
            break;
    }
}

private IHubProtocol hubProtocol;

TeBeCo

@tebeco

i wondered yesterday if there was another attribute like for example

[ColumnSource]
public string Foo => this.HubProtocol.Name;

to generate a column name Foo
but that would also mean finding how to hide the automated one

The attributes are name but as it is attributes value must be known at compile time

i'm very curious here

about inheritance thx, spot that yesterday but not sure this was the only way since it felt weird at first

i did not though about the enum ^^

looks a bit better for this scenario

Adam Sitnik

@adamsitnik

I've forgotten about the simplest solution....

You could just increase the column width

because it's contaisn the type name (the default .ToString() behaviour in .NET), but it's trimmed

TeBeCo

@tebeco

yep looks like typeof(T).FullName

Adam Sitnik

@adamsitnik

dotnet/BenchmarkDotNet#1265

TeBeCo

@tebeco

(.ToString() as you said ^^)

Adam Sitnik

@adamsitnik

I am before my first coffee :d

TeBeCo

@tebeco

any idea for the other tricky question ^^ ?

with or without a coffee ☕️ hahaha
this one is a mystery for me

i saw that there were .WithNuget helpers but my guess is that it's helps full when thing still builds
i'm looking at few breaking change in my own implementation ;)

one way would be to release both version of my code in release and bench against my own Nugets package since i just consume an IHubProtocol ?

Adam Sitnik

@adamsitnik

You could also add a custom build configuration

as example: https://github.com/dotnet/efcore/blob/12e02637ec5fe69048e296ae215abd905b641298/benchmark/EFCore.SqlServer.Benchmarks/EFCore.SqlServer.Benchmarks.csproj#L14-L28

https://github.com/dotnet/efcore/pull/14624#issuecomment-461319553

TeBeCo

@tebeco

i though about that too, it's not a simple breaking change in my own lib, i'll try to see how it goes but it might complexify stuff just for the cost of a bench

while previous version of this lib where already shipped but the exposed "contract" is still the same, the .WithNuget might do the trick if i consume a "local" RestoreSource

i'll just make sure PackOnBuild is enabled to consume the "nuget feed" versus local feed

the breaking change is for MessagePack in my scenario so it's not a simple one ;)

Collin J. Sutton

@insaneinside

Hey folks, new user here. Trying to benchmark something that (in setup) requires creation of a resource, whose location isn't known before it's created but needs to be communicated to the benchmark methods somehow. Is this (a) not possible, (b) a bad idea/wrong way to do things, (c) an indication that I'm using the wrong benchmarking library, or (d) all of the above?

Adam Sitnik

@adamsitnik

hi @insaneinside you are fine ;) what kind of resource are we talking about?

Collin J. Sutton

@insaneinside

@adamsitnik , I'm specifically comparing how a filesystem search heuristic scales depending on certain settings. My plan was to create a set of test directories somewhere in %TEMP%, and compare performance between settings, using [Params("scaledirname1", "scaledirname2", "scaledirname3")] etc, but still need to communicate the root directory for each scale directory. (Yes, I'm aware of the fact that all sorts of things will affect benchmarks based on disk I/O, and am not yet worrying about it.)

Adam Sitnik

@adamsitnik

@insaneinside in such case, most probably the simplest solution would be to create the corresponding directories in the [GlobalSeutp] method based on the value of [Params]

an alternative is to use env vars to pass something to the process

Albert Hives

@ahives

Hi, is it possible to run multiple benchmark classes in the same Program.cs?

Tomasz Kajetan Stańczak

@tkstanczak

hi, is it normal that after selecting a benchmark in the console from BenchmarkSwitcher all the other jobs are executed later

also, is it possible to disable warnings?

liek this -> The minimum observed iteration time is 1.9000 us which is very smal
l. It's recommended to increase it.

Tomasz Kajetan Stańczak

@tkstanczak

the case here is that I know (more or less) what I am doing - I am trying to benchmark memory only and I do not care about the times

Tomasz Kajetan Stańczak

@tkstanczak

I know I can contribute :) but in case I can also make a comment that maybe you can place the table after the warnings and after the legend so there would be no need to scroll up after the benchmark run

the legend is only needed for the users at the beginning, warnings too, and then you often would run a benchmark and want to see the results after it finishes

I end up scrolling up and looking for tables after each run

Stephan Grein

@stephanmg

hey folks.

is there any update on the possibility to use BenchmarkDotNet with Unity?

Where communities thrive

People

Repo info

Activity