Stefano Doni

826 posts

Stefano Doni banner
Stefano Doni

Stefano Doni

@stef3a

AI-driven performance optimization. CTO @akamaslabs https://t.co/VG5qnztkON

Milan, Italy Katılım Ocak 2012
567 Takip Edilen446 Takipçiler
Stefano Doni retweetledi
KubeFM
KubeFM@K8sFM·
🗣️ @stef3a breaks down why Kubernetes optimization initiatives fail. He identifies two critical barriers: technical complexity of configuring parameters and organizational misalignment between teams with conflicting incentives Watch the interview: ku.bz/rdbv-kvWt
English
0
4
2
637
Stefano Doni retweetledi
KubeFM
KubeFM@K8sFM·
🗣️ @stef3a announces the launch of @akamaslabs Insights, a new platform module for Kubernetes teams to optimize performance and efficiency of Kubernetes applications Watch the interview: ku.bz/rdbv-kvWt Read the announcement: ku.bz/mlYTxPC6x
English
0
2
1
709
Stefano Doni
Stefano Doni@stef3a·
@richardstartin So the Datadog agent leverages JFR for data collection? Cool! How does it do that - continous recording on a file, or streaming? It doesn't need to restart the JVM I guess
English
1
0
0
233
Richard Startin
Richard Startin@richardstartin·
@stef3a I’m not familiar with that event but you may be able to tweak the sampling frequency with jdk.ContainerCPUUsage#period 100 ms in a jfp file (googleable setup). There’s examples in the Datadog agent here for config. github.com/DataDog/dd-tra…
English
1
0
2
73
Stefano Doni
Stefano Doni@stef3a·
Hey JVM folks, I'm using JFR to study JVM startup performance. Super convenient, everything in one place. However, CPU usage is rounded to 1s granularity, which is too coarse. Same with jfr print. Any way I can sub-second data?
Stefano Doni tweet media
English
2
0
0
170
Erik Gahlin
Erik Gahlin@ErikGahlin·
@stef3a You can set the sampling period for the event on the command line. See -XX:StartFlightRecording:help Regarding the jfr tool, an exact mode was added in JDK 25, but it currently only works when you use the print command: $ jfr print --exact --events ContainerCPUUsage <file>
English
1
1
3
145
Stefano Doni
Stefano Doni@stef3a·
@richardstartin Hi Richard, thx! Good point about overhead, I'll make sure to check with and without. I don't need profiling or thread-level data, I'm using the ContainerCPUUsage event as total CPU usage is enough for my use case, but I need it at finer granularity
English
1
0
0
202
Richard Startin
Richard Startin@richardstartin·
@stef3a I’d be careful using JFR for startup because it has quite heavy initialisation. I also assume you’re using the thread cpu time events which are very coarse. Try using async-profiler which will give you stacktrace samples based on CPU time per thread, at configurable granularity.
English
2
0
1
39
jholusa
jholusa@jholusa·
Proud to be back to #Current25! 🎉 This time we'll deep dive how #JVM affects the @ApacheFlink runtime. Join me at my talk on Wednesday, 29 Oct 2025 4:45 pm - 5:30 pm CDT 🍻
jholusa tweet media
English
1
2
2
257
Stefano Doni retweetledi
Akamas
Akamas@AkamasLabs·
Optimize Kubernetes environments with Akamas & SpeedScale! 🚀 SpeedScale captures real production traffic while Akamas runs AI-driven experiments for continuous optimization. Say goodbye to guesswork! Full blog post 👇 akamas.io/events/java-on…
English
0
2
1
106
Stefano Doni retweetledi
Bruno Borges
Bruno Borges@brunoborges·
Next week, @stef3a and I will be talking about Java on Kubernetes, Performance Challenges and Solutions. We would love to hear from you with this survey: ➡️ docs.google.com/forms/d/e/1FAI… Just a few minutes of your time.
Bruno Borges tweet media
English
1
1
8
881
Stefano Doni
Stefano Doni@stef3a·
Join me and the awesome @brunoborges to learn the secrets of JVM performance on K8s! How people configure Java & K8s, JVM horizontal vs vertical scaling, K8s requests&limits, biggest config mistakes people do, future trends of JVM & K8s perf, and more!
Bruno Borges@brunoborges

#Java on #Kubernetes: Lessons in Performance Engineering with Akamas and Microsoft. Free webinar with myself and @stef3a --- on September 25th | 9am PT. Register at ---> akamas.io/events/java-on…

English
0
1
5
1.5K
Stefano Doni retweetledi
Charles 🎉 Frye @ GTC
Charles 🎉 Frye @ GTC@charles_irl·
I like making GPUs go brrt at @modal. I wrote up what I've learned along the way in an extension to the GPU Glossary -- our "CUDA Docs for Humans". Introducing: the GPU 𝔓𝔢𝔯𝔣𝔬𝔯𝔪𝔞𝔫𝔠𝔢 Glossary. modal.com/gpu-glossary/p…
English
39
129
954
156.9K
Stefano Doni retweetledi
Go
Go@golang·
“Container-aware GOMAXPROCS” by Michael Pratt and Carlos Amedee — go.dev/blog/container… #golang
English
1
62
287
27.2K
Stefano Doni retweetledi
Akamas
Akamas@AkamasLabs·
Wrong node sizes in #Kubernetes will either squeeze your pods or waste 💰 Autoscaling needs tuning to avoid “stranded capacity.” 🛠️ Choose the right instance types, monitor workloads, and cut costs. 👉 Learn how Akamas can help: hubs.li/Q03BWX-90
English
0
1
1
52
Stefano Doni retweetledi
Volker Simonis
Volker Simonis@volker_simonis·
In the past, people without JVM knowledge tuned JVMs based on random data from the Internet. Nowadays, people without JVM knowledge and no understanding of LLMs tune JVMs based on recommendations from LLMs which were trained on the same random data. Is this progress?
Walldorf, Deutschland 🇩🇪 English
5
9
59
4.1K

Keşfet

@akamaslabs @richardstartin @ErikGahlin @jholusa @ApacheFlink @nearyd @brunoborges @AkamasLabs