Intel Makes Continuous Profiler Open-Source To Help Improve CPU Performance

[phoronix]

Phoronix: Intel Makes Continuous Profiler Open-Source To Help Improve CPU Performance

Intel this morning released Continuous Profiler as open-source, a software solution developed by Intel Granulate for aiming to help boost CPU performance...

https://www.phoronix.com/news/Intel-Continuous-Profiler-OSS

[Kjell]

This would me amazing for Graphics Cards!

I find GPUVIS cumbersome to work with. Are there any better alternatives?

Here's some that I came across with a quick search:

• This one is really interesting: UMR: https://gitlab.freedesktop.org/tomst...ge_requests/37
• RGP: https://gpuopen.com/rgp/
• Perfetto
• Gnome FlameGraphs

[panikal]

This is *cloud native* meaning you store your output data in the cloud, and quite complex for simple CPU analysis. This is meant to be run across multiple services in real time to analyze bottlenecks across all your things. It requires AWS, and uses both S3 and SQS.. I suspect for avg devs, this is waaay too complex and over the top compared to other analyzers and will quickly run beyond AWS 'free tier' sort of use, and enterprises are still grasping at what performance in the cloud means beyond monthly budget costs, if they even care about that beyond 'does it bring a profit'.

Its open source because Intel couldn't find a path to monetize this beast is my guess. I mean it seems like cool tech but it sooo niche for who would actually need this. I can think of a few places I've worked at/with that possibly could have used this, but none that would have benefited from its use, except one, who is already doing similar things. Maybe I'm missing some killer feature here......

[Phoronos]

This is *cloud native* meaning you store your output data in the cloud, and quite complex for simple CPU analysis. This is meant to be run across multiple services in real time to analyze bottlenecks across all your things. It requires AWS, and uses both S3 and SQS.. I suspect for avg devs, this is waaay too complex and over the top compared to other analyzers and will quickly run beyond AWS 'free tier' sort of use, and enterprises are still grasping at what performance in the cloud means beyond monthly budget costs, if they even care about that beyond 'does it bring a profit'.

Its open source because Intel couldn't find a path to monetize this beast is my guess. I mean it seems like cool tech but it sooo niche for who would actually need this. I can think of a few places I've worked at/with that possibly could have used this, but none that would have benefited from its use, except one, who is already doing similar things. Maybe I'm missing some killer feature here......

All wrong.
It is not complex at all.
And it is not open source for not monetizing but for improving the drivers easier and faster.
Stop lying.

[panikal]

All wrong.
It is for GPUs, not CPUs.
And it is not complex at all.
And it is not open source for not monetizing but for improving the drivers easier and faster.
Stop lying.

New troll account? Everything on both their blog and GitHub readme page mention CPU. This is about cloud services though it works with hostname and kubernetes services... My monetization comment was a bit pessimistic but what can I say, I'm an ex Intel Blue badge.

[Phoronos]

New troll account? Everything on both their blog and GitHub readme page mention CPU. This is about cloud services though it works with hostname and kubernetes services... My monetization comment was a bit pessimistic but what can I say, I'm an ex Intel Blue badge.

Yes I meant CPU, not GPU...Anyway, that doesnt change what I said.

[panikal]

What is wrong with what I said? I'll try again. It's a project that's measuring CPU and CPU alone across cloud native services in a way that supports a hybrid DC solution using kubernetes and storage is hard requirement to S3/SQS. Did I miss something?

Its something you can get with an ELK stack (open source) or New Relic (not at all) or anything in that realm but in both of those cases, a LOT more detail tracing API calls and logs and even things like cloudflare response timings or etc... Yes Complex is a matter of perspective. In every case you have 99% of the "complexity" burden post-install. There's nothing in this about drivers you couldn't get much easier - this tool is about distributed services. (edit: yes you can get CPU timings to function level like this in both elk and new relic).

For enterprise that can sustain managing this infrastructure and it's configuration without creating tech debt, more power to ya. Kudos to Intel for releasing this at all I guess, and maybe it will turn into something that's not requiring aws one day.

[Phoronos]

What is wrong with what I said? I'll try again. It's a project that's measuring CPU and CPU alone across cloud native services in a way that supports a hybrid DC solution using kubernetes and storage is hard requirement to S3/SQS. Did I miss something?

Its something you can get with an ELK stack (open source) or New Relic (not at all) or anything in that realm but in both of those cases, a LOT more detail tracing API calls and logs and even things like cloudflare response timings or etc... Yes Complex is a matter of perspective. In every case you have 99% of the "complexity" burden post-install. There's nothing in this about drivers you couldn't get much easier - this tool is about distributed services. (edit: yes you can get CPU timings to function level like this in both elk and new relic).

For enterprise that can sustain managing this infrastructure and it's configuration without creating tech debt, more power to ya. Kudos to Intel for releasing this at all I guess, and maybe it will turn into something that's not requiring aws one day.

Ok, I will try again.
There is nothing wrong with AWS or infrastructure or configuration.
There are many similar projects with continuous profiling, like GWP , Datadog , CodeGuru, Dynatrace, etc
Continuous profiling is widely used in the industry.
Only you see it as a bad thing. Which it is not.