RFR: 8293017: Improve hash calculation parallelism in jmod/jlink

Alan Bateman alanb at openjdk.org
Mon Aug 29 09:22:09 UTC 2022


On Mon, 29 Aug 2022 08:55:06 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> `jmod`/`jlink` are executed during build time to produce the `jmod` and the base modules image. On slow hardware (Raspberry Pi -class, for example) and/or slow VMs (debug, interpreter-only, for example) this takes a while. Profiling shows the considerable time is spent on hashing modules either for writing out the ModuleHash attribute or for verifying the hashes are good.
> 
> Those paths can be parallelized to make them quite faster.
> 
> The major contributors to module hashing are `java.base`, `jdk.desktop` and `jdk.localedata`, so we have a significant opportunity for parallelism here. 
> 
> Motivational improvements on `make clean-images images`:
> 
> 
> # x86_64 Server, release
> 
> # Baseline
> real	0m11.895s
> user	1m4.526s
> sys	0m10.715s
> 
> # Patched
> real	0m10.701s   ; <--- 1.1x improvement
> user	1m5.097s
> sys	0m11.260s
> 
> # x86_64 Zero, release
> 
> # Baseline
> real	5m20.335s
> user	7m7.791s
> sys	0m7.258s
> 
> # Patched
> real	2m23.551s  ; <--- 2.23x improvement
> user	7m14.442s
> sys	0m7.856s
> 
> 
> Additional testing: 
>   - [x] Linux x86_64 fastdebug, `java/util/jar`
>   - [x] Linux x86_64 fastdebug, `tools/jmod`
>   - [x] Linux x86_64 fastdebug, `tools/jlink`
>   - [x] Linux x86_64 fastdebug `tier1`

This issue will require discussion as it potentially impacts usage at run-time when the resolver runs at startup or when creating module layers. I also wonder if it has an impact on JDK builds where many jobs are running jmod at the same time. It may be that this has to be configured by a system property so that it can be enabled only in the JDK build.

Also just to say that Claes removed all the lambda expressions from this code in JDK 9 to help with startup as that will have to checking too. Also if we are changing this code then we should try to keep the style/conventions/line-lengths consistent where possible.

-------------

PR: https://git.openjdk.org/jdk/pull/10060


More information about the core-libs-dev mailing list