Openjdk-infrastructure: Get a more powerful ARM 32 box

[karianna]

[gdams]

@vielmetti is working on sorting out ARM32 on Packet

[vielmetti]

In particular, the ARM32 boxes we have report that they are ARM64 machines but they also run ARM32 binaries, so getting your build environment just exactly right might require some finesse. I am talking to folks who have been through this so that I can give you some guidance, look for something next week if all goes well.

# uname -p
aarch64

[tellison]

Any progress here? I see we are configuring up a new ARM32 machine at the moment, but I believe it is not any "more powerful" than the earlier machine.

[gdams]

Hopefully @vielmetti is going to be able to help us out with more powerful arm32 hardware but it is currently not ready for release in packet from what I understand

[ghost]

Sorry for the silly question but would it help to build OpenJDK for 32-bit machines? I see no OpenJDK binary build for 32-bit Intel and 32-bit ARM GNU Linux, Apple OS X and Microsoft Windows.

[karianna]

@gouessej I think the only viable 32-bit target is ARM these days. I doubt anyone is running a 32-bit intel...

[ghost]

@karianna Of course, as time goes by, less and less people are running 32-bit Intel but numerous libraries still provide 32-bit binaries and Oracle still provide 32-bit JREs:
http://www.oracle.com/technetwork/java/javase/downloads/jre8-downloads-2133155.html

I'd like to use AdoptOpenJDK instead of some old binary builds made by alexkasko whose project is discontinued. I see that ojdkbuild provides 32-bit builds too:
https://github.com/ojdkbuild/ojdkbuild

Is it really a big deal?

[karianna]

It depends on whether we can get 32 bit hosts. We rely on the generosity of IaaS providers :-)

[JakubVanek]

Hmmm, I wonder if running armhf Docker containers on aarch64 would do it. Alternatively if the only needed thing is to build the JDK, then cross compilation could work. However it is a compromise as this doesn't allow for runnings tests natively (outside of QEMU).

ARMv7 Docker images are luckily available from Docker Hub: https://hub.docker.com/r/arm32v7/

[vielmetti]

A brief update in advance of the TSC meeting.

1. There are currently two server-class architectures that support 32-bit code in addition to 64-bit code on Arm servers. The Hisilicon "Hi1616" chip and the Ampere "eMag 1" chip both do this today.
2. Server-class chips from other vendors, including Marvell ("ThunderX" and "ThunderX2") and Qualcomm ("Centriq 2400"), do not have a 32-bit native execution environment.
3. Packet has both Hisilicon and Ampere systems in inventory in very small quantities. When additional units free up, I plan to get Adopt OpenJDK at least one of them and hopefully one of each of them at least for comparison.
4. The long-term availability of server-class devices that support both 32-bit and 64-bit work is uncertain. We expect more vendors to drop 32-bit support for their next generations of high end chips, but who knows, and there's always surprises.
5. The long-term demand for single-board and small devices for 32-bit work is unlikely to diminish, especially as the Android ecosystem still has a whole lot of 32-bit code in it.

Happy to talk through any or all of these points in suitable detail.

[karianna]

Thanks for the update that's very useful detail

[vielmetti]

New 32-bit ready Arm gear is in, so this can start to move forward. I'll coordinate with @gdams @sxa555 to get a machine reserved to the right place so we can test and set up.

[aahlenst]

More powerful ARM32 machines have been ordered.