Hi Dave, > Peter Korn wrote: >> Hi Dave, gang, >> >>> As we all know, the live CD has grown quite a bit since the 2008.05 >>> release. I've spent some time analyzing how we're using the space and >>> where the growth between build 86 and build 94 occurred, and worked >>> up some recommendations. >>> >>> The analysis data is now posted at: >>> >>> http://www.opensolaris.org/os/project/indiana/cd_space/ >>> >>> The recommendations are posted at: >>> >>> http://www.opensolaris.org/os/project/indiana/cd_recommendations/ >>> >>> I would especially appreciate feedback from the desktop, X, and l10n >>> teams on the recommendations. We'll work up an actual plan once >>> we've discussed a bit. >> >> As noted by the analysis data, it may prove difficult to remove >> SUNWgnome-a11y-libs. Looking at the rest of the accessibility >> packages (11MB), I suspect most of the 8MB Dasher package is from the >> very large set of language models (including such critical languages >> as Klingon :-) that we certainly don't need on a non-global CD. >> >> For LiveCD accessible install support, the most significant use case >> is unassisted installation by blind users. GOK & Dasher users are >> more likely to have assistance with OS installation - their >> disabilities are typically severe enough that these users generally >> need assistance with overall computer setup, connecting cables, >> inserting media, etc. Blind users by contrast typically don't need >> assistance with computer setup/install, and we should avoid >> introducing a requirement for sighted assistance in OpenSolaris for >> these users. Especially if the cost in space is only 1.5 - 2.4MB >> (less than 0.4% of the CD). >> > > Thanks for your feedback, Peter. Just to be clear about the > suggestion I made, in case you're not familiar with the product > structure, we presently ship two different live CD's. One has only > the 14 primary languages, while the second one (the global CD) > contains all of those plus the rest of the languages, and uses > different (and slower) compression technology to allow all of that to > fit. This proposal relates only to the primary languages CD. An > unstated assumption (which I probably should have stated, though it > too is up for negotiation, I guess) is that we will continue to ship > those two CD versions because we will not be able to spend time on > improving the global CD's performance before 2008.11. The global CD > still has a fair bit of room on it; we might apply some of the > suggestions to it as well for consistency, but there's currently no > requirement to do so.
I appreciate the distinction, and the time pressure. Given the fact that there is no guarantee that we will continue to ship a global LiveCD, that the accessibility bits we can remove (especially for blind LiveCD installation support) are small, and that we haven't finished evaluating all of the options for removing things we don't need to ship, I would very much like us to keep the option of shipping the primary languages LiveCD with accessibility support on the table. >> Looking at the analysis and goals (to shrink by ~70MB), the CD >> Recommendations page estimates 45MB savings from splitting 32- & >> 64-bit kernel & libraries, plus another 25MB savings from fonts >> (together totaling the 70MB goal). Remove another 6MB for include >> files, plus the analyzed "redundant" packages for another 36MB, and >> you've saved 112MB - significantly in excess of the 70MB target. If >> after all that, more > > In reality, each avenue has some cost inequality, of course, and some > may not be achievable in time for 2008.11, so we might have to take > some expedient and less palatable choices for the short term. Also, > some have drawbacks. For example, splitting 32- and 64-bit can make > it harder for users to obtain a CD that works on their platform, and > potentially complicates stocking and distribution. We chose to split > along languages because we think that's an easier choice for users to > make - they almost certainly know what languages they want to use and > that will be consistent, while having both 32- and 64-bit platforms in > their environment is highly likely (especially since virtualization > technologies like VirtualBox emulate only 32-bit presently) and thus > the distro download size effectively doubles. I'm personally not in > favor of this one because of that issue; I've included it as a > potential choice since I knew it would be brought up. Understood. It's great to see the other e-mails in this thread with thoughtful suggestions for ways we can trim space. Again I come back to how little we save by pulling accessibility - especially (again) the support for blind users. >> space is still needed, we could save another 8MB by dropping Dasher >> and 0.6MB by dropping GOK. I hope this will leave more than enough >> room to keep the 10-11.4MB needed for blind accessibility on the Live >> CD, while still having lots of room left over for driver growth, etc. >> I also note that while the other distros don't include all of the >> accessibility functionality on their Live CDs, Ubuntu at least >> includes blind accessibility with theirs. In fact, that inclusion is >> one of the main reason why Ubuntu is one of the most popular distros >> among blind users on the Orca mailing list. We should at least do >> that much. That said, my preference would be to also keep GOK & >> Dasher for the main languages if possible. If my guess is correct, >> Dasher for the languages supported on the LiveCD won't be very large >> (though it'll take a little work to separate out core language from >> the remainder). >> > > Yes, Will has continually reminded me of Ubuntu's popularity here ;-) > I'd appreciate some help in estimating the size reduction possible in > Dasher, that would be a useful option for us to consider. Will also > suggested I've mistakenly included SUNWspeex here, though if anyone > can comment on that I'd appreciate it. I'm not sufficiently familiar with how packages are put together to give you a solid estimate, but... Dasher installs into /usr/bin/, /usr/share/gnome/help, /usr/share/omf, and /usr/share. The language support is in /usr/share/dasher (all of the "alphabet* and "training*" files). Collecting all of this together, we have an installed size of ~22MB for Dasher. If we strip out the non-English "alphabet*" and "training*" files, and a couple other things that are obviously not needed for English use, we drop to an installed size of 3.2MB. File-by-file gzipped, this minimal collection of files is 975k. 931k if you gzip the tar file. If you can point me to the list of 14 primary languages, I can refine the estimate. But it looks like we should be able to trim most of the 8MB of a Dasher for the primary languages LiveCD. Also - what compression technologies do we use for these two LiveCDs? Regards, Peter Korn Accessibility Architect & Principal Engineer, Sun Microsystems, Inc. _______________________________________________ indiana-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/indiana-discuss
