Doug Leavitt wrote: > I'm curious, is IPS sophisticated enough that we can set up > say a os2008.11 system with zones for 2008.05, 2008.11, > 2008.current [where current updates every 2 weeks, until 2009.xx] > such that we could run the build engine in all three zones > and feed into into 2008.05 to see if it builds, then 2008.11, then ... > until we find a successful build? > > I'm assuming that we would have to use pkg to update each of the > zones that they got fixes, so we could stay in sync with fixes. > Has IPS been tested in that environment yet? > > Also, are there plans on improving pkg.depotd's performance?
Actually, quite a bit of work on performance has been done, but there is still a lot to do. The current depot is about 15-20% faster than the first implementation we had. At the moment though, we're more focused on functionality, correctness, and memory footprint. > pkgsend currently takes ~30 hours to push 10,000 packages > to a depot. Scaling that up, such as 30,000 packages [10k x3 builds] > would be 3-4 days, if we had to do something like a full rebuild > of 30,000 packages x 3 builds [say becasue of a libc fix] that would > be 9+ days, kjust to push... > > 150k x 3 packages would be something like 45 days I think. The performance of your push is also heavily dependent on what method you're using to publish. For example, if you're doing a "normal" set of ips manifest actions, that should be quicker than using svr4 imports, etc. If you can provide some detailed specifics about how you publish each package, along with specific timings, that would be quite helpful. While this sort of performance isn't our primary focus at the moment, we'll gladly take specific bug reports and try to address them as we can. Cheers, -- Shawn Walker _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
