The new version invalidates the cache. So the first install will create a lot more nuget traffic. Is the second install faster?

No, all installs are equally slow.

Ok. That's unfortunate. I'll investigate tomorrow.
On Dec 30, 2014 3:33 PM, "Vasily Kirichenko" [email protected]
wrote:

No, all installs are equally slow.

—
Reply to this email directly or view it on GitHub
#487 (comment).

Do you see this kind of performance issue also on open source project? I think I need something to run the profiler.

Also: how many projects / references files do you have?

And I forgot to mention: we redefined the install process. Now it performs a selective update. This means you can edit the dependencies file or references files and "install" will try to make it work with the minimal changes in the lockfile. In order to do so it runs the resolver.
So it is expected to be a bot slower (maybe not that slow). If you just want to restore the packages, there is also the "restore" command.

That said: the first shot for this selective update is not really optimized and I will try to make it faster. Maybe we could also skip the resolver if we detect no changes. Or things like that.

/cc @haf @agross

I've used it about 30 times today and I think it's really fast. But right now I'm only working towards the public nuget server.

If it's something to do with multiple nuget sources there's already #331 for that.

36 seconds is not fast enough.
Yes, the vast majority of that time is taken by "exploring" our private nuget TeamCity server.

This is our paket.dependencies. The solution contains ~180 projects.

source https://www.nuget.org/api/v2
source http://private/guestAuth/app/nuget/v1/FeedService.svc

nuget AutoFixture
nuget bootstrap 3.3.1
nuget BouncyCastle 1.7.0
nuget CassandraCSharpDriver
nuget Castle.FactorySupportFacility
nuget CommonServiceLocator 1.0.0.0
nuget DotNetZip
nuget EasyNetQ.Management.Client
nuget ExtCore
nuget FAKE
nuget Fizzler
nuget Fizzler.Systems.HtmlAgilityPack
nuget FluentNHibernate
nuget FontAwesome
nuget Foq
nuget FParsec
nuget FsCheck
nuget FsCheck.NUnit
nuget FSharp.Collections.ParallelSeq
nuget FSharp.Configuration
nuget FSharp.Control.Reactive
nuget FSharp.Data
nuget FSharp.Data.SqlClient
nuget FSharp.Management
nuget FSharpx.Core
nuget FSharpx.TypeProviders.Regex
nuget FsPickler
nuget FsUnit
nuget FsWebAddRegistryEntries 0.1.0.0
nuget fszmq
nuget HtmlAgilityPack 1.4.6
nuget Http.fs
nuget jquery.TypeScript.DefinitelyTyped 1.6.0
nuget knockout.TypeScript.DefinitelyTyped 0.6.7
nuget knockoutjs 3.2.0
nuget Microsoft.AspNet.Mvc 5.2.2
nuget Microsoft.AspNet.Web.Optimization 1.1.3
nuget Microsoft.AspNet.WebApi 5.0.0
nuget Moment.js 2.8.4
nuget moment.TypeScript.DefinitelyTyped 0.9.7
nuget mongocsharpdriver 1.9.2
nuget Moq
nuget MSBuild.Extension.Pack 1.3.0
nuget MSDN.FSharpChart.dll
nuget NEST 0.12.0.0
nuget NUnit 2.6.3
nuget NUnit.Runners 2.6.3
nuget Paket
nuget PowerShellTypeProvider
nuget protobuf-net 2.0.0.640
nuget Quartz
nuget RabbitMQ.Client
nuget RegexProvider
nuget RhinoMocks
nuget rickshaw 1.4.6
nuget rickshaw.TypeScript.DefinitelyTyped 0.0.2
nuget Selenium.WebDriver 2.41.0
nuget Selenium.WebDriver.ChromeDriver 2.9.0.1
nuget SevenZipSharp
nuget SQLProvider
nuget SSH.NET
nuget Strathweb.CacheOutput.WebApi2 0.6
nuget Suave
nuget underscore.js 1.7.0
nuget underscore.TypeScript.DefinitelyTyped 0.4.9
nuget UnionArgParser
nuget Unity 2.1.505.2
nuget Unquote
nuget WindowsAzure.Storage 3.0.2.0
nuget ZetaLongPaths
nuget NuGet.CommandLine
  9 private packages

Ok. That's sounds a bit like we broke the request caching.
On Dec 30, 2014 5:02 PM, "Vasily Kirichenko" [email protected]
wrote:

36 seconds is not fast enough.
Yes, the vast majority of that time is taken by "exploring" our private
nuget TeamCity server.

—
Reply to this email directly or view it on GitHub
#487 (comment).

14 seconds if I remove all the private packages. Still too slow.

14 seconds for an install isn't (IMHO) too bad considering it's doing the dependency resolution etc.. Install shouldn't be the standard command to use though - a restore should be much quicker. But if it was 2 seconds before....

Sounds like a good benchmark project ;-)
On Dec 30, 2014 5:07 PM, "Vasily Kirichenko" [email protected]
wrote:

14 seconds if I remove all the private packages. Still too slow.

—
Reply to this email directly or view it on GitHub
#487 (comment).

There is still A LOT room for improvement...

Cool! Two of my projects in there ;)

could you please run the install with 0.20.6?

The second install should be a lot faster - I broke the caching in 0.20.0.

But I still found some other low hanging fruits as a goal for tomorrow.

25 seconds first run, 17 seconds subsequent runs. If I remove all private packages, it works really fast (2 seconds).

Do I understand right that TeamCity support nuget v1 only and it causes the terrible performance?

Awesome.

There is still one improvement I need to do. Caching with more than one
server seems to be still an issue and might decrease perf. I think I can
fix that, too. But not in 2014.
On Dec 31, 2014 5:05 PM, "Vasily Kirichenko" [email protected]
wrote:

25 seconds first run, 17 seconds subsequent runs. If I remove all private
packages, it works really fast (2 seconds).

Do I understand right that TeamCity support nuget v1 only and it causes
the terrible performance?

—
Reply to this email directly or view it on GitHub
#487 (comment).

AFAIK - maybe no longer the case? is that every external NuGet feed causes a perf cost as the whole dependency chain must be searched across all feeds.

Yes that's a separate problem. But here I'm talking about subsequent
install processes (selective update with all versions pinned) without any
changes in the dependencies file or the references files. In theory there
should be no http request in this case. Unfortunately we are still doing a
lot.

If you do a full update it's still probing all feeds.
On Dec 31, 2014 5:21 PM, "Isaac Abraham" [email protected] wrote:

AFAIK - maybe no longer the case? is that every external NuGet feed causes
a perf cost as the whole dependency chain must be searched across all feeds.

—
Reply to this email directly or view it on GitHub
#487 (comment).

please retry with 0.20.10
as before the second install should be a lot faster, even with multiple feeds

32 seconds first paket install (after updating to 0.20.11), 5 seconds subsequent installs.

Great job!

should be even faster now.
Could you please retry with 0.20.21?

4 seconds.