Post AazabCAFbKxpDhZDAe by Profpatsch@mastodon.xyz
 (DIR) More posts by Profpatsch@mastodon.xyz
 (DIR) Post #AazYaSzvKUtFEKGGxs by Ninjatrappeur@social.alternativebit.fr
       2023-10-21T10:23:04.414941Z
       
       0 likes, 0 repeats
       
       Pretty cool talk segment from the GraalVM lead dev where he makes a "ahead of time compilation > just in time compilation" case.He compares several metrics to come up to this conclusion, such as peak throughput, throughput/memory ratio and overall memory usage. He later explains why JIT supposed gains (constant tuning) do not always pay off in practice. That's pretty much in opposition to what I heard for the last 10 years. The guy has been working on a JIT java VM for the last 15 years, I guess he knows what he's saying to some extent.I'm by no means an expert in JITs, I can't determine whether this case makes sense or not yet.More details in the video recording of the talk: https://invidious.fdn.fr/83le8YeejVU?t=1464
       
 (DIR) Post #AazabCAFbKxpDhZDAe by Profpatsch@mastodon.xyz
       2023-10-21T10:45:00Z
       
       0 likes, 0 repeats
       
       @Ninjatrappeur JIT on a good VM can give you superior management in production though, adjusting things and monitoring without a redeploy.
       
 (DIR) Post #AazabD628VBW6ubMEi by Ninjatrappeur@social.alternativebit.fr
       2023-10-21T10:45:37.633610Z
       
       0 likes, 0 repeats
       
       @Profpatsch This is precisely the assumption that's debunked in the talk.
       
 (DIR) Post #AazdBv0CZWxEcEVOUq by Profpatsch@mastodon.xyz
       2023-10-21T11:12:26Z
       
       1 likes, 0 repeats
       
       @Ninjatrappeur great, gonna give it a listen
       
 (DIR) Post #AbX9CLwdJ6ATCOsxlY by hyc@mastodon.social
       2023-11-01T01:37:58Z
       
       1 likes, 0 repeats
       
       @Ninjatrappeur if a JIT is constantly tuning, it must spend more resources on perf monitoring, which detracts from resources available to actually execute the program. And the overhead cost never goes away. AOT compilation can do much deeper analysis if desired, but the cost of analysis is only paid once. Also JIT assumes all measured perf metrics are solely due to the active program, can't account for system load from other processes. It was always a fool's errand.
       
 (DIR) Post #AbX9CQ3W1S39wu8W36 by hyc@mastodon.social
       2023-11-01T03:22:06Z
       
       1 likes, 1 repeats
       
       @Ninjatrappeur JIT optimization looks good on paper, just like Itanium/EPIC/VLIW looked good on paper. They both fail for general purpose computing, because they can't account for other system activity. VLIW still gets some measure of success in GPGPUs because usually all of the programmable hardware is dedicated to a single job at a time. But it's hopeless in a busy multitasking environment. Just like JIT optimization.