 |
 | Home | Contact Us | FAQ | Search & Site Map | Link to Us |
 | Sign In | Join | Other 45 Sites in Network |
Java Forum / General / December 2005Opinions on complexity
I am thinking of developing a network application for my PhD, and it involves (among other stuff) my own implementation of the TCP/IP protocols. So, at the state the Java 1.5 version is, I need the use of JNI for the RAW sockets. (Although the use of Java is not a MUST, I am more experienced at it than I am at C/C++ or other useful programming language.) By now, I've managed to almost-implement the Ping: I send the appropiate ICMP Echo Request to 127.0.0.1 (with its well-formed IP header, ICMP header and data, based on the RFCs), but I still don't "hear" the ICMP Echo Reply from my application, (although I can see this ICMP Echo Reply going back when I capture the network traffic of my loopback interface with Ethereal). But it's not the point of my entry in this group to find a solution on this (it must be something odd in the C native code I have to fix). The thing is that, although I find this project feasible, I don't really know how complex may it become (I am sure the TCP/IP stack is not a trivial thing) and, unfortunately, I cannot spend as much time as I would like on this (you sure know how it works: I've got a job from 8 to 17:30, which is not bad at all, but it doesn't let me too much spare time). So here goes my question: do you think it is a feasible project for developing it on my own in a reasonable time?, do you recommend me "opening" it on sourceforge or something like that, and share it with other thinking heads?, are you one of those thinking heads :) ?. Thanks in advance for your comments, _Hobbes. ____________________________ "I must've put my finger on it." - Calvin [snip] do you recommend me > "opening" it on sourceforge or something like that, and share it with > other thinking heads? IMHO - I do not think using the public for feedback and suggestions justifies YOUR thesis work. This is the point of thesis work. That is, research.. I could be wrong. My thesis work was in a different field, psychology.  SignatureThanks in Advance... IchBin, Pocono Lake, Pa, USA http://weconsultants.servebeer.com/JHackerAppManager __________________________________________________________________________ 'If there is one, Knowledge is the "Fountain of Youth"' -William E. Taylor, Regular Guy (1952-) Yes, IchBin, you're completely right, I wrote the wrong words. I clearly know what I want to do, so I would be the "thinking head" in the project (just a way of saying it) and, what I'm really looking for, are some hands with programming experience, that could follow my guidelines and help me with the programming. So, the idea would be: me making the research and posting in soruceforge the work I have done till now, and people that find it interesting making the programming I ask for. I don't know how efficiently will it work, or even if it is the purpose of a sourceforge-style Web site. > I am thinking of developing a network application for my PhD, and it > involves (among other stuff) my own implementation of the TCP/IP > protocols. So, at the state the Java 1.5 version is, I need the use of > JNI for the RAW sockets. (Although the use of Java is not a MUST, I am > more experienced at it than I am at C/C++ or other useful programming > language.) I'm unsure why you feel the need to re-implement TCP/IP. It's generally implemented fairly well by your operating system. Implementing a toy TCP/IP stack may be within your reach, but a production quality TCP/IP stack is a mess of compromises between various denial of service vulnerabilities, and could take a team of several people several years to get right. (Or it could take less time; because of differences between people in levels of skill and aptitude, which are generally one to two orders of magnitude, no definitive answer can ever be given to a question of that sort.) Signaturewww.designacourse.com The Easiest Way To Train Anyone... Anywhere. Chris Smith - Lead Software Developer/Technical Trainer MindIQ Corporation > I am thinking of developing a network application for my PhD, and it > involves (among other stuff) my own implementation of the TCP/IP > protocols. So, at the state the Java 1.5 version is, I need the use of > JNI for the RAW sockets. (Although the use of Java is not a MUST, I am > more experienced at it than I am at C/C++ or other useful programming > language.) Are you planning to make changes to the TCP/IP protocol? If not, a mere reimplementation of TCP/IP would not be suitable for a PhD thesis. You should pick a unanswered topic in comp-sci and attempt to find a solution. There is no requirement to "optimally solve" the problem but you should be able to show why it doesn't work and propose ideas for others to further the research. Remember: "If I have seen farther than others, it is because I was standing on the shoulders of giants." Sir Issac Newton. SignatureSincerely, Tris Orendorff [Two antennae meet on a roof, fall in love and get married. The ceremony wasn't much, but the reception was excellent.] > Are you planning to make changes to the TCP/IP protocol? If not, a mere > reimplementation of TCP/IP would not be suitable for a PhD thesis. You > should pick a unanswered topic in comp-sci and attempt to find a solution. > There is no requirement to "optimally solve" the problem but you should be > able to show why it doesn't work and propose ideas for others to further > the research. > Remember: "If I have seen farther than others, it is because I was standing > on the shoulders of giants." Sir Issac Newton. I could be wrong, but I'm pretty sure that he has a good idea of what would make a good PhD thesis. =P I think he's just asking about language issues. SignatureOf making better designs there is no end, and much refactoring wearies the body. Wow!, lot of movement here!. I DO like it :). I will try to answer all of you: To: Chris Smith > Implementing a toy > TCP/IP stack may be within your reach, but a production quality TCP/IP [quoted text clipped - 4 lines] > to two orders of magnitude, no definitive answer can ever be given to a > question of that sort.) Well, my initial idea is not developing a commercial TCP/IP stack, but it is not making a toy implementation as well, because I will make extensive use of it for my research. I want to develop a realist implementation. I will not be alone in my way, because some students will start to help me briefly, but they lack the experience I might find somewhere else. > I'm unsure why you feel the need to re-implement TCP/IP. When I first wrote in this forum, I didn't find it necessary to explain my PhD project but, as some of you have asked questions that have to do with it, I will write down a little more about what it intends to deal with: I need to modify some of the values the TCP/IP protocol works with and, the only way of being able to change them is having my own implementation. Hacking a linux kernel and modifying the parameters is not an option, because I need lot of control on what the protocol internally does. So, this answers Tris Orendorff to: > Are you planning to make changes to the TCP/IP protocol? And gives the whole reason to Benji when he says (thanks for entrusting me Benji =D): > I'm pretty sure that he has a good idea of what would make a good PhD thesis To Roedy Green: > Its is designed to work on an almost perfect network. It would > crumble if for some reason only a small fraction of the packets got > through. Long before I had ever heard of TCP/IP I outlined a protocol > designed to survive horrible phone lines that I was dealing with in > the third world. see http://mindprod.com/jgloss/sax.html Thanks for the link, I did not know anything about SAX, but it has a lot to do with my PhD project =) > TCP/IP in itself is not that complicated, but DNS lookup on which it > depends is pretty hairy. This words give me some breath. Technically speaking, I know I can handle it but, again, my main problem is time (I would like to end with it before seeing the seas swallow the continents). So, again, here comes my question: do you think sourceforge (or any other suggested site) could help me with my "time problem"? And, again, thanks a lot to all of you for your comments, _Hobbes >This words give me some breath. Technically speaking, I know I can >handle it but, again, my main problem is time (I would like to end with >it before seeing the seas swallow the continents). Read Comer's book on how TCP/IP works. There is a link to it at http://mindprod.com/jgloss/tcpip.html TCP/IP is peculiar in that the protocol itself is simple, but within its rules, you can be quite clever. TCP/IP is more like a game with simple rules where you can invent strategies to play it well. It leaves a fair bit of latitude for what you can do at each point. It is not like protocols where every "move" is proscribed. SignatureCanadian Mind Products, Roedy Green. http://mindprod.com Java custom programming, consulting and coaching. > Well, my initial idea is not developing a commercial TCP/IP stack, but > it is not making a toy implementation as well, because I will make > extensive use of it for my research. I want to develop a realist > implementation. Is a GC-ed language suitable for a "realistic" implementation ? (It might be, it depends on what /kind/ of realism you expect out of it.) Also, if your purposes are to investigate the performance implications of different design choices, don't you have to start from a baseline that is at least credible by comparison with stacks that have been performance-tuned for years ? (That /is/ a question, btw, not an assertion phrased in rhetorical vein.) > I need to modify some of the values the TCP/IP protocol works with and, > the only way of being able to change them is having my own > implementation. Hacking a linux kernel and modifying the parameters is > not an option, because I need lot of control on what the protocol > internally does. So, in what way will this be of value for anyone (except you) to work on ? OK, intellectual interest, but are you willing to bet your PhD on the kind of person who would find this interesting in the first place (perfectly possible) not finding something even /more/ interesting to play with in a couple of weeks time ? > This words give me some breath. Technically speaking, I know I can > handle it but, again, my main problem is time (I would like to end with > it before seeing the seas swallow the continents). Might be worth scoping it out in (non-optimistic) time plan on the assumptions that you get no help from the OS community /and/ that the students who will (you mentioned) help will, in fact, slow you down more than they help. > So, again, here comes my question: do you think sourceforge (or any > other suggested site) could help me with my "time problem"? My guess is no (for reasons as implied above); but then I'm not an OS enthusiast (nothing against free software -- I both write it and use it -- but the OS community often gives me a pain). Personally I think I'd rather start from an established stack (probably the BSD stack) -- or at least budget two or three weeks to making a concerted effort to understand exactly what I could/couldn't do with that implementation, before committing myself to either a longish development that didn't actually contribute to the PhD, or a high risk strategy of hoping someone else would do (most of) that bit for me. BTW, you might be interested in uIP: http://www.sics.se/~adam/uip/ports.html I don't know anything about that project myself, but the project seems to be active. (Which could be taken as evidence against my OS prejudice). It might also be worthwhile getting in touch with the JITS people: http://www.lifl.fr/RD2P/JITS/TCPIP Again, I don't know anything about it except what a bit of webaging turns up. -- chris Chris Uppal ha escrito: > > Well, my initial idea is not developing a commercial TCP/IP stack, but > > it is not making a toy implementation as well, because I will make [quoted text clipped - 3 lines] > Is a GC-ed language suitable for a "realistic" implementation ? (It might be, > it depends on what /kind/ of realism you expect out of it.) I don't see the relationship between GC (it stands for "Garbage Collection", doesn't it?) and the realism. I am not writing a commercial implementacion of the TCP/IP, but a kind of simulation of it for other purposes. GC will only do its job in what OOP is concerned. Who's going to develop network applications with a Java implementacion of the TCP/IP? > Also, if your purposes are to investigate the performance implications of > different design choices, don't you have to start from a baseline that is at > least credible by comparison with stacks that have been performance-tuned for > years ? (That /is/ a question, btw, not an assertion phrased in rhetorical > vein.) The performance-tuning I want to do is not like the one you are talking about. (Thanks for the acclaration of the "not rhetorical vein", when I started reading your comment I was starting to remember all of your parents :P, but I forgot about them at the end of the acclaration). > > I need to modify some of the values the TCP/IP protocol works with and, > > the only way of being able to change them is having my own [quoted text clipped - 7 lines] > not finding something even /more/ interesting to play with in a couple of weeks > time ? It will have practical value. It won't have a "killer application", but it is not merely intellectual interest. > > This words give me some breath. Technically speaking, I know I can > > handle it but, again, my main problem is time (I would like to end with [quoted text clipped - 3 lines] > that you get no help from the OS community /and/ that the students who will > (you mentioned) help will, in fact, slow you down more than they help. You're right. Thinking about it, I agree with you: they will slow me down more than help. > > So, again, here comes my question: do you think sourceforge (or any > > other suggested site) could help me with my "time problem"? [quoted text clipped - 7 lines] > a longish development that didn't actually contribute to the PhD, or a high > risk strategy of hoping someone else would do (most of) that bit for me. Yes, that was my first thought, but I am not too strong at C, and C++ is not much of my taste. Anyway, I still don't discard that possibility. > BTW, you might be interested in uIP: > [quoted text clipped - 8 lines] > > Again, I don't know anything about it except what a bit of webaging turns up. Thanks for the links! > -- chris >So here goes my question: do you think it is a feasible project for >developing it on my own in a reasonable time?, do you recommend me >"opening" it on sourceforge or something like that, and share it with >other thinking heads?, are you one of those thinking heads :) ?. The problem with Java is it does not give you access to the packet level underlying TCP/IP. TCP/IP from a bits and bytes point of view simple and elegant. See http://mindprod.com/jgloss/tcpip.html You will need JNI for the lowest level. Its is designed to work on an almost perfect network. It would crumble if for some reason only a small fraction of the packets got through. Long before I had ever heard of TCP/IP I outlined a protocol designed to survive horrible phone lines that I was dealing with in the third world. see http://mindprod.com/jgloss/sax.html TCP/IP in itself is not that complicated, but DNS lookup on which it depends is pretty hairy. SignatureCanadian Mind Products, Roedy Green. http://mindprod.com Java custom programming, consulting and coaching. > I am thinking of developing a network application for my PhD, [...] > The thing is that, although I find this project feasible, I don't > really know how complex may it become Isn't that the point of a PhD thesis? To demonstrate that you can do something non-trivial and complex? > (I am sure the TCP/IP stack is > not a trivial thing) Compared to "real" telecommunications protocols it is a piece of cake. Its simplicity is the reason why it go so popular compared e.g. to the once rivaling OSI protocol stack. > do you recommend me > "opening" it on sourceforge or something like that, and share it with > other thinking heads?, are you one of those thinking heads :) ?. Aren't you supposed to do your PhD thesis on your own? Something smells fishy here. What's the name of your university? /Thomas SignatureThe comp.lang.java.gui FAQ: ftp://ftp.cs.uu.nl/pub/NEWS.ANSWERS/computer-lang/java/gui/faq http://www.uni-giessen.de/faq/archiv/computer-lang.java.gui.faq/ Thomas Weidenfeller ha escrito: > > I am thinking of developing a network application for my PhD, > [...] [quoted text clipped - 3 lines] > Isn't that the point of a PhD thesis? To demonstrate that you can do > something non-trivial and complex? Yes, it is, but what I would like to demonstrate with this PhD is NOT that I am capable of developing the TCP/IP stack in Java, but to reach conclusions with its implementation. As I said in a previous message, I need a total control of what the TCP/IP internally does, but the goal is to conclude things based on some tests. It seems that this thread is converting in "what a PhD should demonstrate" or "why the hell do you want to do that". I DO know what I want to do, I'm just asking about the best way of doing it so, please, stop writing comments like this, that drive nowhere. > > (I am sure the TCP/IP stack is > > not a trivial thing) > > Compared to "real" telecommunications protocols it is a piece of cake. > Its simplicity is the reason why it go so popular compared e.g. to the > once rivaling OSI protocol stack. I don't agree. It was pure merchandising: just a race on which one made the best efforts in becoming a standard. TCP/IP is not so simple, it is just a subset of the OSI stack, and both share the same background ideas, which are not trivial. > > do you recommend me > > "opening" it on sourceforge or something like that, and share it with > > other thinking heads?, are you one of those thinking heads :) ?. > > Aren't you supposed to do your PhD thesis on your own? Something smells > fishy here. What's the name of your university? The same as I told you a couple paragraphs before. > /Thomas > -- > The comp.lang.java.gui FAQ: > ftp://ftp.cs.uu.nl/pub/NEWS.ANSWERS/computer-lang/java/gui/faq > http://www.uni-giessen.de/faq/archiv/computer-lang.java.gui.faq/ Thomas Weidenfeller ha escrito: > > I am thinking of developing a network application for my PhD, > [...] [quoted text clipped - 3 lines] > Isn't that the point of a PhD thesis? To demonstrate that you can do > something non-trivial and complex? Yes, it is, but what I would like to demonstrate with this PhD is NOT that I am capable of developing the TCP/IP stack in Java, but to reach conclusions with its implementation. As I said in a previous message, I need a total control of what the TCP/IP internally does, but the goal is to conclude things based on some tests. It seems that this thread is converting in "what a PhD should demonstrate" or "why the hell do you want to do that". I DO know what I want to do, I'm just asking about the best way of doing it so, please, stop writing comments like this, that drive nowhere. > > (I am sure the TCP/IP stack is > > not a trivial thing) > > Compared to "real" telecommunications protocols it is a piece of cake. > Its simplicity is the reason why it go so popular compared e.g. to the > once rivaling OSI protocol stack. I don't agree. It was pure merchandising: just a race on which one made the best efforts in becoming a standard. TCP/IP is not so simple, it is just a subset of the OSI stack, and both share the same background ideas, which are not trivial. > > do you recommend me > > "opening" it on sourceforge or something like that, and share it with > > other thinking heads?, are you one of those thinking heads :) ?. > > Aren't you supposed to do your PhD thesis on your own? Something smells > fishy here. What's the name of your university? The same as I told you a couple paragraphs before. > /Thomas > -- > The comp.lang.java.gui FAQ: > ftp://ftp.cs.uu.nl/pub/NEWS.ANSWERS/computer-lang/java/gui/faq > http://www.uni-giessen.de/faq/archiv/computer-lang.java.gui.faq/ > Isn't that the point of a PhD thesis? To demonstrate that you can do > something non-trivial and complex? Absolutely not. Or rather, it shouldn't be. I'm not familiar with the general standards for giving out the Ph.D. degree by graduate computer science departments. Demonstrating the ability to implement a fairly complex project might be the point of a senior project for a Bachelor's degree. A Ph.D. thesis is intended to demonstrate the ability to contribute theoretically interesting work to a field. Whether the code is complex is immaterial... although clearly it's necessary that the problem itself be non-trivial, which implies a certain level of complexity of the code. Sounds to me like the OP wants a configurable TCP/IP stack to work with, so that he could then perform interesting theoretical work with it. If so, and if that interesting work really is interesting, then I can't see one whit why it would matter to someone evaluating the thesis work how difficult it was to implement the configurable TCP/IP stack. > > do you recommend me > > "opening" it on sourceforge or something like that, and share it with > > other thinking heads?, are you one of those thinking heads :) ?. > > Aren't you supposed to do your PhD thesis on your own? Something smells > fishy here. What's the name of your university? Nothing smells fishy to me about the "supposed to do your own" concern. Lots of open source projects were originally created for research purposes, and some number were explicitly designed to make research easier. One that comes to mind from a Java perspective is Intel's ORP from a few years back, which explicitly rejected optimizations that complicated the JIT and GC interfaces, so as to make it easier for researchers. What does seem unlikely about this particular idea is that the OP will be able to start a working open-source project without contributing the initial code for version 1. My advice here is the same as always. The first step to starting a thriving open source project is to have working code, NOT an idea. Lots of people have ideas. Once you have code that's useful to someone, they will THEN use it and, possibly, come back with enhancements. Signaturewww.designacourse.com The Easiest Way To Train Anyone... Anywhere. Chris Smith - Lead Software Developer/Technical Trainer MindIQ Corporation >>Isn't that the point of a PhD thesis? To demonstrate that you can do >>something non-trivial and complex? > > Absolutely not. Or rather, it shouldn't be. [...] > Demonstrating the ability to implement a fairly complex project might be > the point of a senior project for a Bachelor's degree. I didn't mention "implementation". I just wrote "doing something". The OP sounds as if his "something" is "implementation": OP> I am thinking of developing a network application for my PhD, and it OP> involves (among other stuff) my own implementation of the TCP/IP OP> protocols. If implementing that application (including the TCP/IP stack) is his PhD thesis (your guess is as good as mine ...), then I would expect that he has to do it on his own. /Thomas Thomas Weidenfeller ha escrito: > >>Isn't that the point of a PhD thesis? To demonstrate that you can do > >>something non-trivial and complex? [quoted text clipped - 14 lines] > thesis (your guess is as good as mine ...), then I would expect that he > has to do it on his own. Ummm, yes, it might be misunderstood (as it has been). I'm sorry for the way I started this thread. The thing is that the objective is not the development of the application (as someone has said in this same thread, it would be fine for a computer science project at university, but NOT for a PhD). The objective is the analysis of communication protocols in several ways, and one of the "collateral damages" is the development of an application that helps me understand these protocols, and play a little bit with them. Finally, one of the steps for creating that application, is the implementation of the TCP/IP stack. I chose Java because it is the language I know the best. I am not alone at my PhD: I have a tutor that accepted my proposal of investigation, so I assume it is valuable for a PhD. The only thing I was asking for, was some advice on the TCP/IP stack implementation. My question has finally turned (more or less) in an discussion (constructive in its own way :P) about what a PhD should deal of. Anyway, I think I have extracted some good conclusions, and it has helped me on deciding my next steps. However, any further comments are still welcomed =) > /Thomas > Ummm, yes, it might be misunderstood (as it has been). I'm sorry for > the way I started this thread. The thing is that the objective is [quoted text clipped - 18 lines] > helped me on deciding my next steps. However, any further comments > are still welcomed =) My comment is that I think you're probably better off implementing TCP/IP in C or similar. I know that you know Java better, but it's really not suited to this task. And you can probably find any number of open source TCP/IP implementations in which you can easily tweak the parameters you want to change; I just doubt you'll find them in Java. To be honest, it surprises me to read about a PhD candidate choosing a particular language just because he knows it best. I thought the whole point of getting a CS degree, as opposed to picking up "Learn this language in 21 days" books, is to understand computer science on a deep enough level that you're not tied to any particular language. It makes me wonder if the only reason you're avoiding C/C++ is dealing with memory allocation. And that makes me think you really ought to implement it in one of those languages, because a person with a PhD in CS really ought to at least get some experience with that kind of thing. Especially if his interest is in low-level stuff like communication protocols. Signaturemonique Ask smart questions, get good answers: http://www.catb.org/~esr/faqs/smart-questions.html > Thomas Weidenfeller ha escrito: > [quoted text clipped - 41 lines] > >>/Thomas For what it is worth, I offer my two cents. I assume your analysis of communications protocol, namely TCP over IP, will encompass throughput performance. In this regard, Java's JVM performance will have a big affect on overall communications performance. Perhaps Java, or any interpreted language, is not ideal for this purpose. With that said, and as many have others have said, there are many TCP/IP stacks in c in the public domain. Here is another resource for some micro stacks, and combined with the RFCs for the protocol will speed your development. http://my.execpc.com/~geezer/osd/net/ But, if Java is your choice, you really only need to JNI the low level socket functions that give you access to the interface device, writing you protocol stack in pure java. I would recommend creating a JNI library of bit diddling functions, and even JNI functions to move packet streams to Java variables simply for speed. If you are less familiar with c/c++ than Java, I recommend you code a little as possible in JNI. I assume again you are comfortable with JNI, and you will spend more time coding Java than JNIs. I suspect your JNI library will have less than a dozen native functions. Joseph > Perhaps Java, or any >interpreted language, is not ideal for this purpose. Java is not just an interpreted language. It is interpreted, hot spotted, jitted and AOTed and perhaps a few others I have not heard of. SignatureCanadian Mind Products, Roedy Green. http://mindprod.com Java custom programming, consulting and coaching. >> Perhaps Java, or any >>interpreted language, is not ideal for this purpose. > > Java is not just an interpreted language. It is interpreted, hot > spotted, jitted and AOTed and perhaps a few others I have not heard > of. LOL - Well that is definitely one way to put it -- LTP :) I have implemented a 'toy' IP stack as part of Netsim [1]. The version available online is a little old now, but it's still in very active development. The source is closed, but that is not a fixed situation. The main reason not to release source is so that students don't steal it. I disagree with the reason, but haven't argued much because nobody has expressed real interest in the source. I've considered extending it to send out real packets, and yes, JNI is one option. There are some libraries available to help with this [2] [3], which I presume use JNI. Another option is IPC. You could have a (relatively) simple C program to which your Java program connects, and when you send the C program a packet it will send it to the OS. Of course, either way loses some of the portability that Java tries to offer. [1] www.netsim.info [2] http://sourceforge.net/projects/jpcap/ [3] http://www.jopdesign.com/ejip/index.jsp Free MagazinesGet these publications absolutely FREE for up to 12 months. There are no hidden fees and no obligation. Simply choose a title, complete the application form and submit it. Read more ...
|
Reply to this Message
|
 Sign In
 Join
 My Latest Posts My Monitored Threads My Blog My Photo Gallery My Profile My Homepage Start New Thread Enable EMail Alerts Rate this Thread
|
©2008 Advenet LLC Privacy Policy - Terms of Use
This website includes both content owned or controlled by Advenet as well as content owned or controlled by third parties.