Hi,
I'm trying to pbulk a large subset of pkgsrc for sparc on NetBSD 71.2. Even if I have one of the fastest sparc cpu
(
not to be confused with sparc64 ), a TurboSparc ( 170Mhz ), it's pretty slow. I was wondering how to accelerate this. Distributed build seems to be a viable option however I cannot purchase more of this hardware for now. So I did some search there and there and did some tests. I came to these "architecture"/"setup" options :
- Forget about compiling on sparc. Cross compile everything on faster (amd64) hardware and add some distcc if not enough fast. It seemed to be a good idea but I tried and couldn't make it work properly. Diging on the subject on the internet seems to show that cross compiling pkgsrc is not really functionnal and some say it's still beta. Fell free to explain me you working setup if i'm wrong. (Please dont' refer me to HOWTO-use-crosscompile on ftp.netbsd.org if you haven't tryed it recently)
- Running pbulk on sparc hardware with the help of some amd64 VM/Host running distcc configured to compile binary for sparc. A sparc cross-distcc on amd64 if you prefer. I came to a working setup however it only help when compiling C code. Lot of pbulk is still running on sparc hardware ( awk, perl, etc ...)
- Still run on sparc hardware but with the help of emulated sparc systems (via qemu-system-sparc) and with one or 2 amd64 vm/host for the scan phase ( build_clients and scan_clients in pbulk.conf). All will mount some NFS filesystem for pkgsrc.Complicated setup but it might works.
- i saw there is a "cross-compile=no" in pbulk.conf.This might be an option however if pksrc does not seems to have a functionnal cross compile option, how can pbulk have one ?
I'd like to know your opinon about it. Which setup is better ? Do you have a another setup in mind I haven't though about ?