Chat-20080513
Jump to navigation
Jump to search
This is the archive for the IRC chat on 2008/05/13. Full transcript is below, notes follow.
Participating: Florian, Hagen, Joe, Mathieu, Tom
Major discussion points:
- Liu has been updating the Wiki w/ Quagga notes
- Hagen has been working through some parallelizations basics, passing packets through sockets
- Some discussion about synchronizing w/ Craig's emulation work
- ns-3 demo proposal was submitted to SIGCOMM by Tom
- Joe posted some basic draft code outlining the statistics framework
Transcript
* Now talking on #ns-3 * Topic for #ns-3 is: gsoc: send questions to ns-developers (http://mailman.isi.edu/mailman/listinfo/ns-developers) if no one answers here * Topic for #ns-3 set by mathieu_ at Wed Mar 26 12:09:23 2008 * tomh (i=824c2013@gateway/web/ajax/mibbit.com/x-7f5a787b86ea1c6b) has joined #ns-3 <tjkopena> hi all <tomh> hi joe <tjkopena> hi tom; george mentioned that he wanted to stop by but is at DARPA today and can't join us (not that we had really discussed an organized chat last week) <tomh> yes, who else besides hagen is here for the gsoc chat? <tjkopena> all three are logged in, but I don't know if they're actually here <lacage> morning <tjkopena> hi lacage <hgntmp> as I know fw is finish his master theses these days, not sure if he will join us <hgntmp> good morning to everyone! <tjkopena> hi hagen <tomh> good morning <tomh> I don't know whether LJ__ is joining, but I can give a brief update <tomh> LJ__ has been developing his plan on the wiki: http://www.nsnam.org/wiki/index.php/Real_World_Application_Integration <tomh> Mathieu suggested on friday for him to get started by looking at how to support netlink sockets via the process API <tomh> that is independent of working on the regular socket api which is in flux <tomh> hgntmp, do you have a plan posted like LJ__ has done (on the wiki)? <tjkopena> LJ__'s wiki page looks good <hgntmp> btw: i am _familar_ with netlink sockets for linux (I used this as a transport mechanism some days ago) <tomh> ok, great-- will let him know <hgntmp> but as already said: netlink is one of the biggest protocols in the stack - not that easy <hgntmp> right <tomh> probably we will support, initially, the pieces needed to modify the kernel forwarding table <hgntmp> recently I looked in the olsrd implementation, they use nowadays netlinks socket for linux to manipulate them <hgntmp> (maybe a smooth introduction from userspace side) <fw_> i'm here. <hgntmp> moin <fw_> i sent mail to ns-developers that has the current state of the nsc work <fw_> i suspect that the mail is pending in the moderation queue, it will hopefull hit the list later today <hgntmp> got it <lacage> tomh, hgntmp yes, the goal is to start with only the pieces we need for zebra and these should be relatively minimal. <tomh> it did hit the moderation queue (I have whitelisted your email for the future) <lacage> hgn, so, what is the status of your work on the parallel stuff ? <tjkopena> hgntmp, have you & george been able to hookup & chat specifics? <hgntmp> really similar to the stuff i wrote to you <hgntmp> currently i stopped because I need additional information how I can convert this packet class <hgntmp> the current status: <hgntmp> i start two ns3 processes and both create a socket, one open, bind, listen and accept() <lacage> hgntmp, if you have specific questions for me, it would be more helpful to send them to the ML if you can <lacage> i.e., I am about to answer your email but it would be nice if others could see what you are up to. <hgntmp> the other one connect with the server socket <hgntmp> lacage: ok <hgntmp> afterwards I serialize the packet, get a Buffer instance, PeekData() and write() to the socket <hgntmp> the client, on the other side, read() the data, build a Buffer and afterwards a Packet instance <lacage> out of curiosity, why are you actually marshalling the packets through a socket ? do you want your parallel implementation to work on distributed machines rather than a single multicore machine ? <lacage> (for the latter, you would just need some careful locking, obviously) <hgntmp> lacage: it was the suggestion from george <lacage> ah, ok. <hgntmp> I mean it is still ok, the implementation via shared mem and semaphores (or whatever synchronization is used) could be implemented afterwards <hgntmp> there are some other fundamental challenges in the area of space parallel appraoch <hgntmp> especially for wireless (calculation interferences et cetera) <hgntmp> but I send my proposal to george and we will discuss these challenges in the next couple of days (after the trip) <lacage> cool <lacage> tjkopena, I see that you have started on the ns-3-stat branch. are you going to work on that in the next few days ? <tjkopena> yes <tjkopena> so there's an ok sketch in there <tjkopena> I am going to email vincent & the list later today <tjkopena> this evening I'm going to clean it up a bit, add another simple example (means) and demo hooking to a trace <tjkopena> hopefully I can get vincent involved & work on it together <tjkopena> I only need a couple basic things for now, so hopefully he can pick up on some of the more interesting statistics <tjkopena> I just 1) needed some of this right now and 2) wanted to make sure it was fairly general, and not tied into specific networking metrics <lacage> tjkopena, ok. <tjkopena> if you have comments on that branch I'd be happy to hear them <lacage> fw_, answered you <fw_> lacage: thanks, will read it asap <lacage> tjkopena, I will try to look at it carefully today <tomh> (reposting-- maybe I missed the response) hgntmp, do you have any plans posted like LJ__ has done (on the wiki)? <lacage> fw_, btw, were you able to catch up with sam ? <lacage> fw_, I understand that 8 in the morning is too early for a googler :) <tjkopena> lacage, sounds good; there are some design notes in docs/ as well <hgntmp> tomh: of course - I will do today - I swear! ;) <lacage> tjkopena, yeah, I saw that <lacage> tjkopena, did you have comments on my stats email from last week ? <tjkopena> yes, I thought it was a great writeup <tjkopena> not too sure on the specifics, but the motivations & goals were good, along with the references <tjkopena> I was going to crib most of it when I turn the design notes into a larger writeup <lacage> tjkopena, ok. looks like you are going to write the ns-3-stats paper :) <lacage> tjkopena, btw, did you submit a sigcomm poster ? <tjkopena> I did not in the end <tjkopena> tomh, sorry I did not get back to your email about that <lacage> tjkopena, ok. tom did submit a demo/poster proposal. <tomh> hgntmp, once your work starts to get posted, I've asked Craig Dowell to track it as well, since he is working on real-time scheduler and emulation, and is working some multicore issues there too <lacage> tomh, I replied your last email about sockets <tomh> it would be nice to harmonize craig's emulation work with your work <tjkopena> lacage, I saw that. I may still come out (have to determine that yet), in which case I'd be happy to help w/ the demo & could bring machines (tablets), etc <tomh> tjkopena, yes I submitted a demo proposal-- did you? <tomh> I'll send a copy later <tjkopena> no <lacage> tjkopena, ok. sounds great. <tjkopena> but am still thinking about coming out for it <tjkopena> also, tom (I mentioned this to lacage earlier), I'm coming out to Seattle just before the 4th of July and would be interested in getting together if people have time <tjkopena> I'm buying air tickets to night & will email everyone <tomh> great, it is a nice time around here to visit <lacage> tomh, the beginning of summer :) <lacage> I am still waiting for it... <tomh> lacage, we have two more days of winter then you should be happier <lacage> yeah, I saw the WE forecast <tomh> lacage, I added a SocketsDefault class last night <tomh> if you think there is a cleaner approach, we can look at it <fw_> lacage: yeah <fw_> lacage: already talked to him about nsc integration stuff. <lacage> tomh, as I replied, I think that attributes would work. <hgntmp> I must leave now - for additional info regarding my person I will read the IRC backlog afterwards <lacage> hgntmp, ok. have a good night <hgntmp> or email! ;) <hgntmp> 5:46pm <lacage> fw_, were you able to setup a regular meeting ? or is this just a couple of emails ? <fw_> emails and jabber * hgntmp (i=50f62021@gateway/web/ajax/mibbit.com/x-94ba2ca630b3953c) has left #ns-3 <fw_> at the moment i send him a weekly status report <tomh> lacage, well, I am not going to hold up my post to ns-developers for that (which I'm trying to write in parallel to this chat) <lacage> fw_, ok. if you can and if he is ok with it, it might make sense to send it to ns-dev. <lacage> tomh, certainly. I will comment on the list. <tomh> I have not pulled the raw buffer variants. That will still be controversial despite what we discussed on Friday. I would prefer to let it lie. <lacage> what does "let it lie" means ? <lacage> does it mean that you do not want to do it _now_ ? or that you do not want to do it _ever_ ? * lacage is curious <tomh> not reopen the issue <tomh> I will comment it on the list <lacage> ok <lacage> thanks <tjkopena> ok guys, have to run; will chat later. thx * Disconnected ().