This is the mail archive of the cygwin-xfree@cygwin.com mailing list for the Cygwin XFree86 project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: 30+ reasons why X-Win32 is Better than Cygwin/X


--- Christopher Faylor
<cgf-no-personal-reply-please@cygwin.com> wrote:

> I didn't read much of this email when I first
> responded to it since I
> suspected that I'd feel compelled to respond and it
> would interfere with
> my "real job".

I know the feeling...

>  But, now it's Saturday, so...

And it is Saturdy here, too, so you even get a timely
response;)

> On Thu, Mar 10, 2005 at 12:57:53PM -0800, Matthew
> Johnson wrote:
> >>Cygwin/X installs with the cygwin setup and this
> is fast too.
> >
> >Not always.  Setup.exe presents a list of mirrors
> with no commentary
> >and not even a _hint_ of which is closest to the
> user in the Web.  I
> >tried 5 or 6 different mirrors in that list before
> I could find one
> >that was "fast too".
> >
> >That is pretty poor performance.  RedHat should not
> allow them to
> >mirror Cygwin if the mirror will not provide faster
> response and better
> >connectivity.
> 
> Red Hat does not control the Cygwin mirror list any
> more than Red Hat
> controls Cygwin.

I see. Perhaps tehy should, much as Sun has at least
some limited 'control' over Netbeans, even though that
too is (almost) entirely a volunteer project.

But really, who controls the mirror list is only
secondary at best: whoever it is who is controlling
it, should do some 'quality-control' on the mirror
list.

>  Cygwin is an all-volunteer
> project.

Hopefully, this does not mean that any volunteer can
'volunteer' his contribution even when it is not a
good one;) I have refrained, for example, from
offering my laptop connected on a dial-up line as a
mirror site!

> Here is how it works: Mirror sites volunteer to
> mirror the Cygwin
> release.  The site is added to the mirror list and,
> subsequently, a
> program checks twice a day to make sure that it and
> the other sites are
> up-to-date.  If a site isn't up-to-date it is
> dropped from the list.

Perhaps that progam should check for more than just
'up-to-date'. I think you dismiss that idea too
quickly below.
 
> I can't think of any useful way to determine "faster
> response and better
> connectivity".  If you are in Michigan and you chose
> a mirror in Brazil,
> you'd undoubtedly see poor performance in your
> download.

No doubt. But that is why I also mentioned in my
original email: with many of those mirrors, the
setup.exe user has no idea where the mirror is
physically located (too many names end in 'org'). So
the poor user in Michigan cannot even guess if the
mirror is in Brazil (with a few exceptions).

For that matter, users can't really rely on teh domain
name suffixes for geographical info either:
http://www.sem40.ru, for example, _should_ be in the
FSU, but is really in Israel.

But many other mirror lists _do_ provide a hint (for
figuring out where the server is). This is what Cygwin
should also require of any 'volunteer' who want to
mirror Cygwin. Some indication of the bandwidth the
mirror is capable of would be good too. Again, many
other lists of mirrors on the Internet already do
this.

>  If you are on
> a network that is being subject to a denial of
> service, you'd see poor
> performance.  There are all sorts of factors which
> can impact *your*
> download performance that have nothing to do with
> how well-connected the
> mirror site is.

Yet all these problems you mention have not prevented
other mirror lists from providing much more help to
the user by providing geographical and connectivity
information to the user when he is asked to select a
mirror. Sourceforge, for example, provides the
geographical location of the mirror, and provides a
good spread in the list presented to the user. I
forget which mirrors I saw that indicate bandwidth.

So the fact that denial-of-service attacks are
possible is not a good reason for failing to provide
this help to the user.

> I can imagine some kind of system which tries to
> figure out connectivity
> by checking the output of traceroute or some similar
> utility but I doubt
> it would ever be useful.

I think you are too doubtful. For that matter, if you
think tracert won't do the trick, you could always do
occasional sample downloads and remove from the list a
site that consistently does poorly -- for geography is
no excuse here: connectivity really is that good
(nearly) worldwide now. I can connect from California
to Siberia more easily than I can connect to most of
those mirrors!

The only "hard part" I see in implementing something
like this is deciding the cutoff for 'poorly'. But I
think once you see the statistics, you will find that
this is not as hard as you fear. And you can always
err on the side of being conservative: cut off only
the worst 10%, for example, after a week of testing.
Even that will be a big improvement over today's
situation.

> I've never seen anything
> like this in any of
> the other projects which use mirrors.  If you have a
> pointer to
> something that does this, however, please provide
> it.

But you must have seen at least the geographical
information listed by sourceforge. Even just that
would be a huge improvement over setup.exe as it is
today. And it is an easy change. I like changes that
provide so much bang for the buck, especially when
they don't close the door on even better changes
later. So I hope that whoever maintains setup.exe will
also appreciate the value of this simple change, and
do it.
 
> I assume that most users are like me.  They find a
> mirror which works
> for them and they stick with it.  That's what I do
> with cygwin,
> sourceforge, Fedora, etc.

And many users, I am sure, do just that. But I have
had to support users who gave up because they tried 5
or 6 different mirror sites and _still_ could not do
the initial download of Cygwin within 24 hours. And
since these users ALL live and work in areas that are
very welll connected to the Internet, this is a sure
indication that something is very wrong with these
mirrors.

By contrast, with sourceforge, it doesn't even seem to
matter which mirror I pick: they have all worked fine
for me. I am not asking that Cygwin's mirrors reach
_that_ level of perfection, but I am asking that we
make a budge in that direction, preferably before my
clients rebel!

> All of that said, however, I'm not a huge fan of
> setup.exe.  I think
> it's UI sucks and I wish someone had the time to
> provide something
> better.

Well, if more people don't give up on Cygwin because
their initial attempt to download runs into the sand
(as described above), then maybe we can find people to
do just that;)

Now don't get me wrong: I do think it is unfortunate
that people would give up on Cygwin that easily. But I
know how important first impressions are, and
setup.exe does not make a good first impression: I
suspect the mirror list is the easiest part of
setup.exe to fix to make a better first impression.
 
> >> > 10      Corporate Tech Support  No      Yes
> >> > 11      Corporate Bug Fix Support       No     
> >> Yes
> >
> >[snip]
> >
> >>Is Cygwin/X worth it's money?  Definitly yes *g*
> >
> >No, NOT 'definitly [sic] yes'.  It depends on how
> much your _time_ is
> >worth.  If your time is worth little to you, or you
> already _have_ much
> >expertise with X, Cygwin and Cygwin/X, then yes, it
> is worth it.  But
> >if you cannot afford to lose the time grappling
> with installations that
> >do different things on different machines,
> demanding you rebuild
> >password files but then refusing to let you do it
> etc, or with
> >"community support" that consists of answers so
> terse (and all too
> >often rude) they are harder to understand than the
> original problem
> >etc, then no, it is not worth it.
> 
> Right.  In open source, people who answer your
> questions may be as rude
> or terse as the people who are asking for help.

Or ruder, or terser. I have seen all combinations.

> The
> answerers also may
> be as clueless as many tech support personnel.

That, fortunately, I have seen only rarely;) A more
common problem is that yes, the answerer really does
know something about it, but has made too many
unwarranted assumptions in his answer, so that the
answer is unintelligible to those who needed the help
in the first place.

This is another example of why it is better to err on
the side of verbosity, as long as the verbage _does_
have real semantic comment, AND does not involve
complex compound sentences that are a burden to parse.
Otherwise your interlocutor will either have to post
_again_, or he will simply give up and go back to
native Windoze wasting money on VisualStudio. My
clients _are_ threatening to do just that.

> I do
> think it is pretty
> rare for people who are helping to comment on
> misspellings or bad
> grammar, however.  YMMV.

And my mileage _does_ vary! When the misspelling or
bad grammar does not hinder communication, then I pass
over it in silence (as I have already done in these
Cygwin lists, even in my brief time here). And I see
many in this list do the same. Good! When, however, it
leaves me with a response that purports to be helpful
but has no genuine semantic content at all, then I
have little choice but to comment. And that is what I
did.

Now I admit that my comment (in another thread, on the
need to rebuild passwd/group files) could have been
even gentler than it was, but I do still maintain that
it was not rude.

I have been going on about this for much too long now!
But it is because I really do feel that Cygwin has a
lot to gain from this simple fix, even if my idea for
testing connectivity is ignored. So I am going to stop
now, unless I am asked specific questions I can give
specific answers to.



-----------------------
Matthew Johnson
mej1960@yahoo.com


		
__________________________________ 
Do you Yahoo!? 
Make Yahoo! your home page 
http://www.yahoo.com/r/hs


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]