Status of packages on the sparc64 architecture
MarkLinimon note 20120610: some of the information on this page is stale. It has been noted.
The state of the packages on sparc64 leaves a lot to be desired. The sources for this problem:
- We do not have a large sparc64 user community.
- Few FreeBSD maintainers and committers have access to sparc64 machines.
- The port building cluster for sparc64 is seriously underpowered.
- We really only have enough horsepower to build one set of packages. Even with an incremental set it takes several weeks, as opposed to only a few days for amd64/i386.
- It takes some portmgr overhead to run sparc64 builds; however, most of the overhead is in flagging ports as BROKEN and sending notifications. (The builds, once started, do not generally require attention unless a machine panics or goes offline, or a disk overflows).
Volunteering your machine to the pointyhat build cluster
- To be minimally useful for package builds the machines would need at least 512MB RAM (preferably at least 1GB).
- CPU speed is obviously highly important (but also hard to come by with sparc64 hardware).
- Having remote access is mandatory. This is because it is likely that frequent reboots and/or kernel debugging will be needed. That is most easily accomplished if there is a serial console available remotely.
- The build machines should have at least 20GB, preferably 40GB or so. Much larger amounts of disk will not be efficiently used. To cut down on network traffic, somewhere local to the machine should be an FTP mirror which would be configured to mirror a distfiles tree.
- Machines will be used 24x7, and package building tends to eat hard drives.
Tracking package status
current package build status (all buildenvs)
package build comparison across all buildenvs
- There are 5 Netra 1s in the package build cluster, with a few other machines on loan to us from various people. (One of the netras now has a dead disk.)
- Our biggest need right now is for people to be using and fixing ports, rather than adding more horsepower to the build cluster. (linimon is working on finding out if we can add more 440MHz-class machines; slower ones might not help us much.)
- We need to decide if we have enough momentum to make it worth maintaining the "desktop" kind of apps on sparc64, or whether we would be better advised to drop back to just "server"-oriented applications. We clearly have a significantly larger user base for and64 and especially i386.
Contact MarkLinimon for more information.