hi this is John Galloway and today I'm talking to McCall from the Bing team and McCullough you wrote a really cool blog post recently where you talked about how the performance impact and all kinds of other great impacts from running on dotnet core to one um can you give me just an idea of the the sigh you know the how how busy Bing is I mean how what kind of traffic you're you're processing here yeah I mean Bing comm gets thousands of queries of search queries per second we power Bing calm and other API partners using cognitive services so you get thousands of queries per second and we have millions of lines of c-sharp code that's powering Bing calm Wow okay so I always think it's great when I'm talking to people to get it you know when I'm talking to customers to tell them that large implementations of dotnet or you know people are using dotnet core on large production instances and this is a really cool one um so one thing you call out right at the top of this post is the idea of ready to run images and I don't think that's that well-known can you explain what that is yeah so ready to run images are this version file format that's available on dotnet core basically allows you to pre compile your assemblies into both native and manage code in the same file which allows you to basically have this ready run experience you have managed code that gets jittered usually but when your images are ready to run that are essentially ready to run when your program starts which is very useful for if you have a large application with many assemblies and a lot of managed code it really helps startup performance and also impacts your working set so usually this code that is generated is more compact in its representation because there's not a lot of jitter jitan going on that's putting code in different places so it's all in one image very cool so as opposed to paying the price of like jetting on each of your servers on a farm you can you can kind of do this at Build time and blow the out to everybody is that correct exactly the the key part of ready-to-run images and dotnet core is that we did have previous technologies like engine on dotnet framework but they always a required admin privileges on the actual machine that you need to Engen on and also the fact that it had to be done on the exact machine so if you have five thousand machines that mossis happens five thousand times yep okay so next this is awesome seeing this chart here this you you showed a thirty four percent improvement from from deployment from upgrading to dotnet core to one so previously you had both dotnet framework for seven to and dotnet core to oh so just that that bump two to one was a thirty four percent improvement overall yeah so we basically when we migrated our code we were still running on net four seven two and then you know we would switch back between four seven two in two Oh and so we had both we saw both sides of four seven two and two oh and when we upgrade to two point one that was phenomenal 34 percent improvement that's right so I love how you went through and called out detailed issue or like features and also specific pull requests on some of these and you showed why your performance got better so like any favorites here I mean I I always loved things like vectorization and and looking at how that kind of bubbles all the way up the stack yes I I you know they're all they have the best part in one sentence is the fact that the community and microphone please contributing open-source has really helped us one my favorite I have to say is the vectorization because it actually you know had a major impact but this my my other favorite one is the combination of work done by Microsoft employee and Andy Ayers and an open-source contributor well-known Ben Adams where they had D virtualization and how that impacted when you did it for certain default equality comparison areas that was you know it was a great sort of synergy between Microsoft II's and open-source desolate I would say you know that's one of my favorites and it's it's really neat in as part of this my understanding is that you had specific requests and you work through the public github thing to say hey we'd love this sort of performance improvement or change and then as part of that collaboration this shipped as part of dotnet core to one and helps everyone out right that's right one of the ones that I that I contributed and you know sort of open to pull requests an issue was this one with calling direct and it's it's because we generate that kind of code so you can imagine as an open source if you have a particular kind of code that you'd like optimized it's it's really great you can just go in and contribute and that's what we did with the help of the.net team as well so it really overall was a great experience operating an open-source is probably one of the main reasons why I think this performance improvement across so many different things has been seen anything cool ok so then one final thing is the actual rolling out to production and this is just amazing here that within two days of the release you were able to deploy what what is it about dotnet core that allows you to do that so Donna core has basically these two deployment types one is the shared deployment the shared one time where you will have the dotnet tool available in your Linux distribution or your Windows OS and the other one is this idea of self-contained apps so for us we use this idea of self-contained apps and we're able to basically have the dotnet core runtime be a part of our application deployment so which if you think about it is is extremely cool you just have your app sitting right besides it is all the core CLR the GC the JIT everything there so if you think from a person a perspective of you know how can you isolate things it's really cool you on day one you have this application running on a framework on day two you upgrade it with dotnet core sitting right beside your app and so when dotnet core 2.1 comes out for us it's a change in the CS project literally a change in the NuGet package we made one character change and we were able to deploy dotnet core 2.1 two days after and that two days is just because of some additional validation we had to do so that was really awesome that's so cool I love here I mean the focus of this post is on performance but were there other features you were able to turn on as a result of moving to dotnet core to one yeah so this is another one of these things where the ecosystem really helped we with done at core 2.1 you also get broadly support in the core FX libraries so we actually use that and were able to get big comm broadly support which was phenomenal so not only did we get all these performance improvements our response eyes also shrunk a little bit because of using broadly so that was also pretty phenomenal wow that's great well that's all the time we have for this quick look I'd encourage people to go out and read more on this post the post is titled bing.com runs on dotnet core 2-1 and thank you so much for your time a call thank you
No comments:
Post a Comment