Skip to content

Latest commit

 

History

History
315 lines (236 loc) · 12.9 KB

README.md

File metadata and controls

315 lines (236 loc) · 12.9 KB
Raw

MMTk-OpenJDK

This repository provides binding between MMTk and OpenJDK.

Contents

Requirements

We maintain an up to date list of the prerequisite for building MMTk and its bindings in the mmtk-dev-env repository. Please make sure your dev machine satisfies those prerequisites.

Before you continue

The minimal supported Rust version for MMTk-OpenJDK binding is 1.61.0. Make sure your Rust version is higher than this. We test MMTk-OpenJDK binding with Rust 1.66.1 (as specified in rust-toolchain). You may also need to use ssh-agent to authenticate with github (see here for more info):

$ eval `ssh-agent`
$ ssh-add

Getting Sources (for MMTk and VM)

To work on MMTk binding, we expect you have a directory structure like below. This section gives instructions on how to check out those repositories with the correct version.

Your working directory/
├─ mmtk-openjdk/
│  ├─ openjdk/
│  └─ mmtk/
├─ openjdk/
└─ mmtk-core/ (optional)

Checkout Binding

First, clone this binding repo:

$ git clone https://github.com/mmtk/mmtk-openjdk.git

The binding repo mainly consists of two folders, mmtk and openjdk.

  • mmtk is logically a part of MMTk. It exposes APIs from mmtk-core and implements the VMBinding trait from mmtk-core.
  • openjdk is logically a part of OpenJDK. When we build OpenJDK, we include this folder as if it is a part of the OpenJDK project.

Checkout OpenJDK

You would need our OpenJDK fork which includes the support for a third party heap (like MMTk). We assume you put openjdk as a sibling of mmtk-openjdk. Cargo.toml defines the version of OpenJDK that works with the version of mmtk-openjdk.

Assuming your current working directory is the parent folder of mmtk-openjdk, you can checkout out OpenJDK and the correct version using:

$ git clone https://github.com/mmtk/openjdk.git
$ git -C openjdk checkout `sed -n 's/^openjdk_version.=."\(.*\)"$/\1/p' < mmtk-openjdk/mmtk/Cargo.toml`

Checkout MMTk core (optional)

The MMTk-OpenJDK binding points to a specific version of mmtk-core as defined in Cargo.toml. When you build the binding, cargo will fetch the specified version of mmtk-core. If you would like to use a different version or a local mmtk-core repo, you can checkout mmtk-core to a separate repo and modify the mmtk dependency in Cargo.toml.

For example, you can check out mmtk-core as a sibling of mmtk-openjdk.

$ git clone https://github.com/mmtk/mmtk-core.git

And change the mmtk dependency in Cargo.toml (this assumes you put mmtk-core as a sibling of mmtk-openjdk):

mmtk = { path = "../../mmtk-core" }

Build

Note: MMTk is only tested with the server build variant.

After cloned the OpenJDK repo, cd into the root directiory:

$ cd openjdk

Then select a DEBUG_LEVEL, can be one of release, fastdebug, slowdebug and optimized.

$ # As an example, here we choose to build the release version
$ DEBUG_LEVEL=release

The differences between the four debug levels are:

$DEBUG_LEVEL Debug Info Optimizations Assertions MMTk Cargo-Build Profile
release ✘ ✔ ✘ release
optimized ✘ ✔ ✘ debug
fastdebug ✔ ✔ ✔ debug
slowdebug ✔ ✘ ✔ debug

If you are building for the first time, run the configure script:

$ sh configure --disable-warnings-as-errors --with-debug-level=$DEBUG_LEVEL

Then build OpenJDK (this will build MMTk as well):

$ make CONF=linux-x86_64-normal-server-release THIRD_PARTY_HEAP=$PWD/../mmtk-openjdk/openjdk images

The output jdk is then found at ./build/linux-x86_64-normal-server-release/images/jdk.

Note: The above make command will build the images target, which is a proper release build of OpenJDK. It is essential that you use this target if you are planning on evaluating your build (e.g. measuring performance, gathering minimum heap values, etc). However, if you are simply developing and building incremental changes often, you may want to use the default target or "exploded image", which has a marginally shorter build time. However, be wary, as the exploded image is the (roughly) minimal set of outputs required to run the built JDK and is not guaranteed to run all benchmarks. It may have bloated minimum heap values as well.

The exploded image can be built as follows. The output jdk can be found at ./build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk.

$ make CONF=linux-x86_64-normal-server-$DEBUG_LEVEL THIRD_PARTY_HEAP=$PWD/../mmtk-openjdk/openjdk

Again: do not use the exploded image for performance analysis.

Profile-Guided Optimized Build

In order to get the best performance, we recommend using a profile-guided optimized (PGO) build. Rust supports PGO builds by directly hooking into the LLVM profiling infrastructure. In order to have the correct LLVM tools version, you should install the relevant llvm-tools-preview component using rustup:

$ rustup component add llvm-tools-preview

In this example, we focus on the DaCapo benchmarks and the GenImmix collector. For best results, it is recommended to profile the workload you are interested in measuring. We use fop as it is a relatively small benchmark but also exercises the GC. In order to best tune our GC performance, we use a stress factor of 4 MB in order to trigger more GC events.

First we compile MMTk with profiling support:

$ RUSTFLAGS="-Cprofile-generate=/tmp/$USER/pgo-data" make CONF=linux-x86_64-normal-server-release THIRD_PARTY_HEAP=$PWD/../mmtk-openjdk/openjdk images
$ rm -rf /tmp/$USER/pgo-data/*

We clear the /tmp/$USER/pgo-data directory as during compilation, the JVM we have created is used in a bootstrap process, resulting in profile data being emitted.

We then run fop in order to get some profiling data. Note that your location for the DaCapo benchmarks may be different:

MMTK_PLAN=GenImmix MMTK_STRESS_FACTOR=4194304 MMTK_PRECISE_STRESS=false ./build/linux-x86_64-normal-server-release/images/jdk/bin/java -XX:MetaspaceSize=500M -XX:+DisableExplicitGC -XX:-TieredCompilation -Xcomp -XX:+UseThirdPartyHeap -Xms60M -Xmx60M -jar /usr/share/benchmarks/dacapo/dacapo-evaluation-git-6e411f33.jar -n 5 fop

We have to merge the profiling data into something we can feed into the Rust compiler using llvm-profdata:

$ /opt/rust/toolchains/1.66.1-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llvm-profdata merge -o /tmp/$USER/pgo-data/merged.profdata /tmp/$USER/pgo-data

The location of your version of llvm-profdata may be different to what we have above. Make sure to only use a version of llvm-profdata that matches your Rust version.

Finally, we build a new image using the profiling data as an input:

$ RUSTFLAGS="-Cprofile-use=/tmp/$USER/pgo-data/merged.profdata -Cllvm-args=-pgo-warn-missing-function" make CONF=linux-x86_64-normal-server-release THIRD_PARTY_HEAP=$PWD/../mmtk-openjdk/openjdk images

We now have an OpenJDK build under ./build/linux-x86_64-normal-server-release/images/jdk with MMTk that has been optimized using PGO.

For ease of use, we have provided an example script which does the above in .github/scripts/pgo-build.sh that you may adapt for your purposes. Note that you may have to change the location of llvm-profdata.

Location of Mark-bit

The location of the mark-bit can be specified by the environment variable MMTK_MARK_IN_HEADER. By default, the mark-bit is located on the side (in a side metadata), but by setting the environment variable MMTK_MARK_IN_HEADER=1 while building OpenJDK, we can change its location to be in the object's header:

$ MMTK_MARK_IN_HEADER=1 make CONF=linux-x86_64-normal-server-$DEBUG_LEVEL THIRD_PARTY_HEAP=$PWD/../mmtk-openjdk/openjdk

Valid object bit

To support the vo_bit (valid object bit) feature in mmtk-core, you can set the environment variable MMTK_VO_BIT=1 when building OpenJDK. This will set the feature for mmtk-core, as well as compiling the fastpath for the VO bit.

$ MMTK_VO_BIT=1 make CONF=linux-x86_64-normal-server-$DEBUG_LEVEL THIRD_PARTY_HEAP=$PWD/../mmtk-openjdk/openjdk

Test

Run HelloWorld (without MMTk)

$ cat ./HelloWorld.java
class HelloWorld {
    public static void main(String[] args) {
        System.out.println("Hello World!");
    }
}
$ ./build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/javac HelloWorld.java
$ ./build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java HelloWorld
Hello World!

Run HelloWorld (with MMTk)

Pass -XX:+UseThirdPartyHeap as java command line arguments to enable MMTk.

$ ./build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap HelloWorld

If DEBUG_LEVEL is release, you should just see

Hello World!

If DEBUG_LEVEL has other values (such as slowdebug), you should see logs, too.

[2023-09-14T06:18:46Z INFO  mmtk::memory_manager] Initialized MMTk with GenImmix (DynamicHeapSize(6815744, 8377073664))
[2023-09-14T06:18:47Z INFO  mmtk::util::heap::gc_trigger] [POLL] nursery: Triggering collection (1670/1664 pages)
[2023-09-14T06:18:47Z INFO  mmtk::plan::generational::global] Nursery GC
[2023-09-14T06:18:47Z INFO  mmtk::scheduler::gc_work] End of GC (304/1664 pages, took 19 ms)
[2023-09-14T06:18:47Z INFO  mmtk::util::heap::gc_trigger] [POLL] nursery: Triggering collection (1680/1664 pages)
[2023-09-14T06:18:47Z INFO  mmtk::plan::generational::global] Nursery GC
[2023-09-14T06:18:47Z INFO  mmtk::scheduler::gc_work] End of GC (460/1664 pages, took 11 ms)
[2023-09-14T06:18:47Z INFO  mmtk::util::heap::gc_trigger] [POLL] nursery: Triggering collection (1674/1664 pages)
[2023-09-14T06:18:47Z INFO  mmtk::plan::generational::global] Nursery GC
[2023-09-14T06:18:47Z INFO  mmtk::scheduler::gc_work] End of GC (614/1664 pages, took 11 ms)
Hello World!

Run DaCapo Benchmarks with MMTk

First, fetch DaCapo:

$ wget https://sourceforge.net/projects/dacapobench/files/9.12-bach-MR1/dacapo-9.12-MR1-bach.jar/download -O ./dacapo-9.12-MR1-bach.jar

Run a DaCapo benchmark (e.g. lusearch):

$ ./build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -Xms512M -Xmx512M -jar ./dacapo-9.12-MR1-bach.jar lusearch
Using scaled threading model. 24 processors detected, 24 threads used to drive the workload, in a possible range of [1,64]
===== DaCapo 9.12-MR1 lusearch starting =====
4 query batches completed
8 query batches completed
12 query batches completed
16 query batches completed
20 query batches completed
24 query batches completed
28 query batches completed
32 query batches completed
36 query batches completed
40 query batches completed
44 query batches completed
48 query batches completed
52 query batches completed
56 query batches completed
60 query batches completed
64 query batches completed
===== DaCapo 9.12-MR1 lusearch PASSED in 822 msec =====

MMTk options

MMTk has many options defined in https://github.com/mmtk/mmtk-core/blob/master/src/util/options.rs

You can use environment variables started with MMTK_ to set those options. For example, export MMTK_THREADS=1 will set the number of GC worker threads to one. Follow the link above for more details.

You can also set those options via command line arguments: -XX:ThirdPartyHeapOptions=options, where options is key=value pairs separated by commas (,). For example, -XX:ThirdPartyHeapOptions=stress_factor=1000000,threads=1 will set stress_factor to 1000000, and threads to 1.

Some OpenJDK options are also forwarded to MMTk options.

  • -XX:ParallelGCThreads=n (where n is a number) sets the number of GC worker threads.
    • MMTk option: Options::threads
    • Note that OpenJDK also has an option -XX:ConcGCThreads. As we have not added any concurrent GC plans into mmtk-core yet, that option is ignored when using MMTk.
  • -XX:+UseTransparentHugePages enables transparent huge pages.
    • MMTk option: Options::transparent_hugepages

Options set via command line arguments take prioritiy over environment variables starting with MMTK_. If both the environment variable MMTK_THREADS=1 and the command line argument -XX:ParallelGCThreads=2 are give, the numberof GC worker threads will be 2.