On Wed, Dec 17, 2008 at 6:54 AM, Matthew Toseland
<toad at amphibian.dyndns.org> wrote:
> [17:48:05] <ESR_> Having used this system in a pretty heavy-duty way now I
> have some observations:
> [17:48:51] <toad_> hmmm?
> [17:48:53] <ESR_> 1. Design of the web interface -- very good.  The
> arrangement of the controls is clever and logical.
> [17:49:37] <toad_> hmmm that's an interesting opinion, there are dramatically
> different ones ...
> [17:50:11] <ESR_> I think it's impressive.  Drab-looking, but who gives a
> shit?
> [17:50:17] <toad_> well basically it's still too geeky, and not enough
> bundled - it needs to incorporate more functionality (i.e. Freetalk), and
> geek-centric stuff needs to be on submenus
> [17:51:06] <ESR_> Could use more stuff bundled in, I suppose, but it does what
> it does quite well.
> [17:51:15] <toad_> also status messages can still take up a lot of room - we
> should move the bookmark updates to only live on the browse page, n2ntms to
> live on the friends page etc, with a single notification on the homepage
> [17:51:20] <toad_> okay, what's 2. ?
> [17:52:58] <ESR_> 2. Performance efficiency of the tools, awful.  My machine
> has been brough to its knees by *two uploads*.  I can't imagine what the Java
> could be doing to eat my proocessor this way, but X is very sluggish.
> [17:53:16] <toad_> yeah most likely that's due to running out of memory
> causing constant garbage collection
> [17:53:23] <evanbd> Looking over logs and the docs on the web site...  I think
> the problems ESR had / is having may have as much to do with poor and
> outdated documentation.
> [17:53:32] <toad_> that's a common problem, the db4o branch solves it at the
> cost of some more disk i/o
> [17:53:34] <ESR_> sorry, that should have been "*two processes*".
> [17:54:03] <toad_> (and slightly more non-GC CPU usage)
> [17:54:45] <ESR_> evanbd: Well, of course!
> [17:54:48] <toad_> the db4o branch should be merged in the next two months, it
> won't solve every performance problem, but it should solve those related to
> memory usage and excessive garbage collection
> [17:55:18] <toad_> evanbd: imho a well designed UI doesn't need lots of
> documentation, nobody reads the docs until they run into trouble ...
> [17:55:51] <ESR_> toad_: You're right.  Documentation is an admission of UI
> design failure.
> [17:56:15] <toad_> so for example it should be possible, and easy, to insert a
> freesite from within the web interface
> [17:56:20] <evanbd> Yes, but for one reason or another Freenet has shown a
> persistent inability to build such a UI.  I think this merits consideration
> of either the reason that is happening or alternate fixes.
> [17:56:29] <ESR_> (This is probably my most shicking heresy to a Unix
> audience :-) )
> [17:56:36] <toad_> ESR_: :)
> [17:56:54] <ESR_> Aaargh,,,can't type.
> [17:57:02] <toad_> evanbd: maybe constant firefighting has something to do
> with it? :|
> [17:57:11] <toad_> evanbd: i should have merged the db4o branch a month ago :|
> [17:57:29] <evanbd> I'm sure it does.  I'm not saying those reasons are bad
> reasons, merely that they appear to exist.
> [17:57:59] <flaushy> toad_: i would disagree with the statement about
> documentation, but i do think, that that is a field where community can do a
> charme
> [17:58:09] <ESR_> 3. Props to you guys for being smart and responsive and
> really caring about your code.  The quality of the on-channel support here is
> high.
> [17:58:11] <toad_> but we do have some idea of what needs to change, in the
> bug tracker and elsewhere ... ian thinks we may be able to get a (specific)
> sponsor to pay for a GUI redesign, we'll see whether anything happens to that
> [17:58:32] <toad_> ESR_: thanks
> [18:00:41] --> mpp has joined this channel (n=mpp at i538776A4.versanet.de).
> [18:00:46] <ESR_> Your guess about the low memory limit leading to excessive
> GC and VM thrashing seems plausible now I've heard it.
> [18:00:59] <toad_> that's what it's been in the past imho
> [18:01:06] <toad_> not in every case but in many cases
> [18:01:32] <toad_> CPU-intensive stuff runs on low-priority threads, so it
> ought not to have a big impact on responsiveness, although we will push your
> load avg up
> [18:02:20] <ESR_> With 17 uploads done the problems seems to be gone, so I
> think number of simultaneous uploads may be a driver; it's probably not often
> anyone pushes that as hard as I just did.
> [18:02:28] <toad_> yep
> [18:02:38] <toad_> well when they do they've learned to up the memory
> limits :|
> [18:02:39] <DuClare> less segfaulted :(
> [18:03:15] <ESR_> The real question is why you have a memory limit at all.
> You're using a language with GC...

Talking about GC. I always wonder why we have to free buckets manually.

When i did the TempBucket finalizer, I log the leak with MINOR, but you think
this is a bug and should be logged as error
(see 
http://archives.freenetproject.org/message/20081125.165832.9d917f92.pt-BR.html
)

The later attempts to fix the bug by freeing manually caused some other bugs..
(see https://bugs.freenetproject.org/view.php?id=2775 )

In fact, after r23847, the leak can be collected by GC.
IMO they don't need any fixing.

> [18:03:34] <toad_> perhaps ... java always has a memory limit afaik
> [18:04:00] <toad_> and GC and swapping go together *REALLY* badly
> [18:04:09] <toad_> so it makes a certain amount of sense
> [18:04:14] <evanbd> Yes, it does.  There's no reason you *couldn't* set it
> large, but it might be better not to.
> [18:06:13] <ESR_> 4. Recovery from failure conditions is really good, ans
> something you don't explain well enough.  J. Random user is likely, foer
> example, to miss the fact that uploads go in a persistent queue ansd *aren't*
> aborted if you kill jsite or close the web interface.
> [18:06:37] <toad_> jsite uploads *ARE* aborted if you kill jsite
> [18:06:51] <toad_> it's one major deficiency in jsite cos it means it's
> impossible to insert huge sites
> [18:06:56] <ESR_> I'm sorryt, I was think thaw.
> [18:06:59] <toad_> ah
> [18:07:05] <ESR_> I weas thinking of thaw.
> [18:07:25] <toad_> yeah ... thaw is sort of third party ... it's great but we
> don't bundle it cos it's not part of the base UI ...
>
> _______________________________________________
> Devl mailing list
> Devl at freenetproject.org
> http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
>

Reply via email to