Sadiq Jaffer

. @avsm and @sadiqj have been telling me about their work on geotessera. Interesting to see embedding requests from universities and organisations around the world, from studying Indonesia's mangroves to making better maps of Greenland's west coast. github.com/ucam-eo/geotes….

I am excited to have this opportunity to work more closely with the amazing people in the @CCI_Cambridge and @cambridge_uccri philanthropy.cam.ac.uk/impact-of-givi…

A cool new feature OCaml 5 version is the ability to continuously monitor production runs through low-cost "runtime event tracing" added by @sadiqj. This should enable remarkable insights into the runtime behaviour of OCaml programs.

Amazing, @sadiqj just landed a huge step towards multicore into OCaml trunk with the safepoints changes #event-5000840157" target="_blank" rel="nofollow noopener">github.com/ocaml/ocaml/pu…

@yminsky I'm not sure what the current JVM GCs do during marking but there certainly is prefetching ahead of the allocation pointer. I did some digging here: opsian.com/blog/jvms-allo… Experiments to see how effective that trick is in OCaml has been on my TODO list for a while.

I'm especially curious about the prefetching improvements. Is this something that other GCs (Go, Java, Haskell, Python, JavaScript) already have?

How much more low-hanging fruit is there in the OCaml GC? First best-fit, now, prefetching. Marking time reduced by 1/3-2/3!?! github.com/ocaml/ocaml/pu…

First blog post in our new Profiling in Production series. Learn what profiling is and how it works: buff.ly/3ooq7Ro #performance #profiling #jvm

Find out how Opsian helped TransferWise fix an elusive performance problem buff.ly/2VXywP2 #performance #java #jvm

Great to finally announce this. Have had customers running it for over six months now and who have had huge platform-wide performance wins.

@RichardWarburto nos habla de "Continuous Profiling in Production: What, Why and How" 👉youtube.com/watch?v=R-1Uyv… #JConfPeru2020

Big thanks to @shipilev for his suggestions and feedback on this one

New article on why Java's Thread-Local Allocation Buffers are so important and why write contention on multicore is probably worse than you might expect: opsian.com/blog/jvm-tlabs… #java #performance #openjdk

@shipilev Glad you are back home - wishing you a swift recovery.

@nipafx Thanks for inviting me, was a pleasure and I learnt a fair bit. Till next time!

Thanks again for being there, @sadiqj! I learned a lot about bytecode, the JVM, and the way this feature is implemented. Next time, we're gonna have some actual numbers to present. :)

Looking forward to speaking on "Production Profiling: What, Why and How" @allthetalksconf tomorrow - 19:45 London time.

After a not very successful first stab at benchmarking NPE performance in #Java14, I got help for a second try: On Thursday, 23rd, at 1900 UTC, Opsian's @sadiqj will join me on stream and together we'll dig into Java 14's most thrilling feature (who said "records"? 🤨).

@immad The Culture series! Persevere through Consider Phlebas - it sets the scene. The next three (Player of Games, Use of Weapons and Excession) are amazing.

Any one got any good space based scifi book recommendations? Latest ones I read that I liked: - The Murderbot Diaries by Martha Wells - The quantum thief by Hannu Rajaniemi - Red Rising Trilogy by Pierce Brown

Continuous Profiling of a JVM application in Kubernetes #java #performance #kubernetes #profiling #jvm buff.ly/2UeyW3g

@acmel @brendangregg Indeed - it's a great resource.

@sadiqj Are you aware of @brendangregg 's brendangregg.com/perf.html ?

Have been doing some low-level optimisation recently and discovered a few neat Linux perf tricks that are worth sharing because they were incredibly useful and helped reduce latency in some parts of the system by 10x.

@acmel Looking forward to stitching lbr. Lbr was useful in cases where the dwarf information was broken - an unfortunate hazard of hacking on a compiler.

@sadiqj You can try with "--call-graph lbr", this method will be greatly improved soon, see lkml.org/lkml/2019/11/1…

@alblue I think we were capturing both at the time, restricting to just user traces is a good idea we should try again.