2011.09.08 Weekly Check in - mattwigway/OpenTripPlanner GitHub Wiki
13:29 <novalis_> Good afternoon, everyone
13:29 <demory_> hello all
13:30 <demory_> shall we get started?
13:30 <andrewbyrd> hi
13:31 <andrewbyrd> sure
13:31 <novalis_> I'm ready
13:31 <demory_> well we've normally been starting w/ the trimet stuff -- Frank, you there?
13:32 <demory_> otherwise there's other stuff we can cover
13:32 <demory_> Frank
13:33 <demory_> FrankP_: sorry, just let us know when you've joined
13:33 <FrankP_> hey David, I'm here
13:33 <demory_> oh, hey. great
13:34 <demory_> so, let's do a quick update on the beta.. i don't have much to report.
13:34 <demory_> i know there are issues w/ i.e. etc
13:34 <novalis_> I've been working with the interns on various routing issues.
13:34 <demory_> for the elevation stuff. i've just been focused on this demo for next week
13:35 <demory_> i will try to set aside some time this weekend for elevation stuff -- both i.e. bugs and the graph builder stuff
13:36 <demory_> sorry novalis_ -- anything of particular note to discuss regarding routing?
13:37 <novalis_> Well, we have a couple of hairy issues: #477 and #492
13:37 <novalis_> 492 is confusing to me; I thought I understood it, but I do not.
13:37 <novalis_> I've asked Andrew to help look into it
13:38 <novalis_> #477 is not at all confusing; it's a straightforward consequence of how we do street splitting for network linking
13:38 <novalis_> But the fix is quite hairy; if anyone cares, I can post the diagram I sent Andrew.
13:38 <andrewbyrd> novalis_ : right, I'm planning on seeing what I can do about these over the next couple of days
13:38 <novalis_> Cool, awesome
13:38 <FrankP_> My update: working on the UI the past few days, directions and alerts. Also working on OTP cluster behind IP-loadbalancer (I'm seeing OTP sometimes get hung for unknown reasons in test, and we also have the slow trips, so I'm a bit concerned there).
13:38 <demory_> great
13:39 <andrewbyrd> did you see the email I just sent you?
13:39 <novalis_> I hadn't yet
13:40 <novalis_> andrewbyrd, that seems much easier.
13:40 <novalis_> And we can test it out at basically no cost.
13:40 <andrewbyrd> novalis_ : and as a general approach it can solve some other quirks as well
13:42 <andrewbyrd> to fill you all in, the routing engine finds ingenious ways to cheat and get around restrictions (turn restrictions in the ticket case, restrictions on sequences of trips in my case)
13:42 <novalis_> OK, then #492 is still confusing, but we have a plan on #477
13:42 <andrewbyrd> by taking shortcuts through stops and boarding/alighting trips
13:44 <demory_> ok. i haven't looked at 492 so I can't add much there -- maybe after next Thurs
13:44 <andrewbyrd> I won't go into detail I guess but refusing to traverse classes of edges in a certain order is equivalent to more complicated arrangements of vertices
13:45 <andrewbyrd> I will still plan on looking at #492 then. I haven't yet reproduced and observed it.
13:45 <demory_> great, thanks
13:46 <demory_> anything more on 477/492?
13:46 <demory_> if not -- FrankP_ going back to your update, is this hanging a new thing?
13:46 <andrewbyrd> nope, I think we have a starting point for addressing them
13:46 -!- kpw [43ddb10f@gateway/web/freenode/ip.67.221.177.15] has quit [Ping timeout: 252 seconds]
13:47 <FrankP_> demory_, hard to say if it's new...it's just that folks (inters) are testing new things. In their testing, they've gotten OTP into a thread-lock a couple times.
13:48 <demory_> hmmm
13:48 <FrankP_> When I restart OTP, and run their test URLs, OTP is fine.
13:48 <novalis_> I wonder if there's a garbage collection issue.
13:49 <demory_> ok well keep us posted on that
13:50 <demory_> again, i'll have more time to look into other stuff after next week
13:50 <FrankP_> I did hook jvisualvm up to the remote server, and saw the trips that take a long time #352 have a staircase memory profile, which may indicate a memory leak after GC's, novalis_
13:52 <novalis_> OK, so that's on the list to investigate
13:52 -!- kpw [43ddb10f@gateway/web/freenode/ip.67.221.177.15] has joined #opentripplanner
13:53 <demory_> ok so we have a couple things to look into
13:53 <andrewbyrd> For these occasional very slow response times, which are generally due to slow secondary searches after a first itinerary has been found
13:53 <FrankP_> If I find something repeatable (beyond #352), I'll ticket it...
13:53 <andrewbyrd> as a stopgap fix GenericAStar does have a timeout option
13:54 <FrankP_> andrewbyrd, timeout option...interesting. Do tell how I can activate that.
13:54 <andrewbyrd> this is of course not a fix, but having it in place could save you some grief during the beta
13:54 <FrankP_> yup...
13:54 <demory_> yeah that would be better than nothing, esp if one itin has already been returned
13:55 <FrankP_> Is there a way to turn that on via .xml config files?
13:55 <andrewbyrd> currently it is a timeout on individual searches, but the path/routing services could progressively lower the timeout as itineraries are returned to bail out when things got out of control
13:56 <novalis_> +1
13:56 <andrewbyrd> often the end user would just see two itineraries instead of 3
13:56 <andrewbyrd> and Frank could receive a log message telling us which trips were causing problems
13:57 <novalis_> andrewbyrd, would Brian's multiple itinerary method avoid this issue?
13:57 <andrewbyrd> I don't think it's configurable via XML now but it could be done
13:58 <FrankP_> being about to control this would be cool. I'd probably set the timeout for 10-15 seconds maximum...the UI will time out in 30 seconds if no itinerary is back, so they get nothing now anyway.. And getting a log would help a lot.
13:58 <FrankP_> Shall I ticket this?
13:58 <FrankP_> (And honestly, I don't need XML control, if this was just default behavior).
13:59 <andrewbyrd> I have been trying out Brian's method (to be fair to Brian, it's my method inspired by a discussion with him, so his in OBA may be more refined)
14:00 <demory_> FrankP_ I think ticketing this is a good idea
14:00 <demory_> ok well it sounds like we have strategies of some kind for the major issues.. anything more on Portland?
14:01 <FrankP_> nothing more here...
14:01 <novalis_> nor for me
14:01 <andrewbyrd> It does work, but at least for the benchmark test cases is not faster than trip banning and is prone to producing technically correct but counterintuitive results
14:02 <novalis_> It's not faster even in the worst cases?
14:02 <andrewbyrd> unfortunately it is slower in the worst cases
14:03 <novalis_> Alas.
14:03 <novalis_> Well, we can put that on the back burner for now
14:04 <andrewbyrd> and this is in the version where I'm progressively hashing tripIds so there's very little overhead from tracking trip sequences
14:05 <demory_> agree w/ novalis_, it sounds like this probably isn't a realistic option for the Oct launch
14:05 <demory_> but def somthing to keep looking into long term. thanks andrewbyrd for the update
14:06 <andrewbyrd> right, not for oct.
14:06 <demory_> well i can talk briefly about our presentation next week
14:06 <demory_> Kevin and I are planning basically a four part presentation:
14:06 <demory_> 1. general project history/background
14:07 <demory_> 2. the Portland launch with demo of new features (bike triangle, improved narrative, etc)
14:07 <demory_> 3. the data management tools we're developing for Atlanta
14:08 <demory_> and 4. the analytics extension and isochrone test
14:08 <demory_> I don't have much to add on #4 beyond the email I sent to the dev list earlier
14:09 <demory_> but we should have something basic to show next week
14:09 <demory_> i'll mainly be working on performance over the next few days. it really bogs down now over about 60 min max weight
14:10 <kpw> but not because of otp, correct?
14:10 <demory_> correct
14:10 <kpw> that's primarily due to rendering of the result set
14:11 <demory_> it's generating the polygon from the set of visited vertices
14:11 <kpw> i'm curious if we can have otp generate a better structured result set that only includes "most distant" points
14:11 <demory_> i'm using a pretty crude approach w/ JTS multipoints right now
14:11 <demory_> the problem is that there are many redundant points
14:12 <demory_> i'm sure there is an efficient way to do this, I just need to think this through some more
14:13 <demory_> anyway we can keep that discussion going on the dev list
14:13 <demory_> oh i did have one other thing about this
14:13 <demory_> right now i'm working within the OTP codebase (in a branch) but I'd like split this into a freestanding project
14:14 <demory_> like we described last week
14:14 <FrankP_> +1
14:14 <demory_> has anything happened w/ putting the current core OTP modules on a maven repo somewhere?
14:15 <demory_> that's what we discussed right?
14:15 <andrewbyrd> what I've done is generate or choose a set of sample vertices (a grid, via decimation, etc.) and only display or save weights from those
14:15 <novalis_> That has not happened and it is my fault for not giving Andrew access to repo.opengeo until today
14:16 <demory_> no worries, it's not high priority
14:16 <andrewbyrd> well, I didn't send you an email asking for access until today either
14:16 <andrewbyrd> OTP is basically already set up to do this
14:16 <demory_> but once it's ready i will try to break out the analyst code
14:17 <demory_> but I won't do anything there until Thurs at the earliest
14:17 <demory_> well, if folks have any other input on the presentation, let me or Kevin know
14:18 <demory_> and I'll keep you updated as we finalize the materials over the next couple days
14:18 <kpw> also, curious to hear thoguhts on who else we should be talking with
14:18 <kpw> this fall we'll be doing a "road show" of sorts for analyitics and we're making a list of interested parties/applications
14:19 <kpw> just be thinking about ti
14:19 <kpw> it
14:19 <FrankP_> Speaking of talking...I might not be around next week. I'll be in Dever, FOSS4G.
14:19 <kpw> great! say hi to the opengeo crew for us
14:19 <demory_> yeah I had that for the end but let's discuss now
14:20 <demory_> Kevin and I will be St Pete on Thurs too
14:20 <demory_> our talk is that morning
14:20 <demory_> should we just cancel the call? or try to reschedule?
14:20 <novalis_> I am fine either way
14:21 <demory_> if we reschedule it would have to be Fri for me
14:21 <FrankP_> cancel makes sense for me, I won't be back till in the office till the 19th
14:21 <demory_> ok let's cancel then
14:22 <FrankP_> that said, I can hang out on irc
14:22 <demory_> sure. but we won't do a call then
14:23 <novalis_> OK, cool.
14:23 <demory_> alright, the only other thing I had was this letter re the European Commission contest
14:24 <demory_> i've been working today on the draft that was started a month or so back
14:24 <demory_> i'm hoping to have something more polished before I leave NYC this afternoon
14:25 <demory_> so the draft was written under Andrew's and Wojciech's names
14:25 <demory_> is that still our plan?
14:26 <demory_> we'll want to touch base w/ W too before we send
14:26 <demory_> but if we're going to send something it should probably go out tomorrow
14:27 <andrewbyrd> I guess we should leave my and Wojciech's names on it since we are the ones in Europe
14:28 <demory_> ok. well I'll let both of you know once I've finished w/ my edits, hopefully within a couple hours
14:28 <demory_> that's my main thing for this afternoon once the chat is finished
14:29 <demory_> Andrew are you ok w/ actually sending it?
14:30 <andrewbyrd> Yes, I will send it. Electronic version?
14:30 <demory_> that's what I figured
14:31 <demory_> i guess we could send a hardcopy too, though I would still do electronic so that they have something in hand tomorrow
14:31 <andrewbyrd> OK, thanks for working on that. If you need to run today and want me to clean it up later, just let me know.
14:31 <demory_> great, thanks
14:32 <andrewbyrd> The page I am looking at states that the deadline is the 15th
14:32 <demory_> really? i thought it was the 9th?
14:32 <andrewbyrd> 15th of october even
14:32 <demory_> i mean, it doesn't matter that much since we're not doing a full submittal
14:34 <demory_> now, if you open the .doc file under step 1 it still says 9/9
14:35 <demory_> but I guess what is on the web is the most recent?
14:35 <andrewbyrd> well we can just do it now if you have time
14:36 <demory_> i may just send a quick email to that address to confirm
14:37 <demory_> if it is in fact Oct 15, that could change things -- perhaps we'd be able to get the Pacific NW demo to the point where we could do a full submittal
14:37 <andrewbyrd> true
14:37 <demory_> let me follow up by email
14:37 <demory_> and if it's in fact tomorrow, we'll just do the letter
14:37 <demory_> it looks like they pushed it back though