From kitabi at usc.edu Mon May 2 15:35:19 2005 From: kitabi at usc.edu (lubaina kitabi) Date: Mon May 2 15:35:56 2005 Subject: [Csci551-talk] GHT question Message-ID: Hello, In the GHT paper, in the section that talks about structured replication , i was just wondering if replication is a misnomer. From what I could gather in that part was that structured replcation is used to create mirror nodes so that the home/root node does not become a hotspot. That would imply that for the same key value, the home node would have some data while the remaining data would be distributed among the "mirrors". Thus , the data is not being "replicated " just "distributed". Please correct me if I am wrong. Thank you, Lubaina Kitabi. From johnh at ISI.EDU Mon May 2 18:26:10 2005 From: johnh at ISI.EDU (John Heidemann) Date: Mon May 2 18:26:50 2005 Subject: [Csci551-talk] Re:Security in Freebits In-Reply-To: Message-ID: <200505030126.j431QAh8032717@dash.isi.edu> On Thu, 14 Apr 2005 13:08:45 PDT, nikhil bhatia wrote: >Sophia, >I was not expecting any replies on my quest. Nice to see your reply :) > >I am reffering to a hypothetical situation. manager.conf if just to simulate the feature we indend to have in our p2p system. >We could also have a freeloader spawned separately which somehow knows the tracker ip/port. Well essentially we could have >a "freeloader.conf" which would spawn freeloader in the network.These freeloader essentially could be "high energy" nodes more efficient than other nodes. > >Well another idea, we could have energy as a parameter. Which will reduce every communication/computation iteration >(sum of packets->sent, packets->rcvd, we could also ignore computation cost as comm costs form majority of enerfy consuption) For peer-to-peer protocols, energy is not usually a concern (except perhaps for GHT). Freeloaders are a concern, though. There is some research in this area; typically they use approaches like reputation systems to judge who is and isn't good. -John Heidemann From johnh at ISI.EDU Mon May 2 20:04:53 2005 From: johnh at ISI.EDU (John Heidemann) Date: Mon May 2 20:05:51 2005 Subject: [Csci551-talk] Ideal Freeloader In-Reply-To: <4272A043.6030609@usc.edu> Message-ID: <200505030304.j4334rMQ002568@dash.isi.edu> On Fri, 29 Apr 2005 13:59:47 PDT, Aneeket Patil wrote: > Whenever a node acts as a freeloader.........ideally in a real world >scenario......he/she never announces that it has the file to the tracker >right? >So ideally no node should even send the request for a file to a freeloader. > I just wanted to confirm , that in Project B , we are supossed to >request the freeloader........and log the response. >We are doing this only to observe the effects of freeloaders ,right? > So, In an real world scenario , the freeloader never announces that it >has the file to the tracker , right? >Aneeket In the real world I believe a freeloader must announce interest in the file to the tracker to find out about other peers. Project B and Project A do not try to exactly mimmic real bittorrent, however. -John Heidemann From rmchandr at usc.edu Tue May 3 00:32:12 2005 From: rmchandr at usc.edu (rashmi chandrasekhar) Date: Tue May 3 00:33:05 2005 Subject: [Csci551-talk] GHT question Message-ID: I think data is initially replicated. But later as queries for events are generated, all mirror nodes are queried, so, it is sort of like distributed. Below is what I could gather from the paper. What they are saying is when a hash of a key maps to a geographic location, all nodes will try to send all data to that node at that location (even if it has to traverse the entire perimeter in doing so) for storage (and queries too). If there are many events happening with that key (suppose weather forecast for every minute) this is when the home node becomes a hot spot. To avoid this, the initial home node data is mirrored at suitable nodes as described in the paper. Now, when any node in the network detects a packet (or rather event) that hashes to the home node, instead of doing a perimeter forwarding and sending it to the home node all the way round the perimeter, the node just sends it to the nearest mirror. The nearest mirror then consumes this packet. This way load is well balanced. When later queries arise in the network, they have to be sent to all the mirrors however, because each of them could have stored different event packets forwarded to them from their neigh bors. If I am wrong in my understanding of this concept, please let me know. -Rashmi. ----- Original Message ----- From: lubaina kitabi Date: Monday, May 2, 2005 2:35 pm Subject: [Csci551-talk] GHT question > Hello, > In the GHT paper, in the section that talks about structured > replication , i was just wondering if replication is a misnomer. > From what I could gather in that part was that structured > replcation is used to create mirror nodes so that the home/root > node does not become a hotspot. That would imply that for the same > key value, the home node would have some data while the remaining > data would be distributed among the "mirrors". Thus , the data is > not being "replicated " just "distributed". > Please correct me if I am wrong. > > Thank you, > Lubaina Kitabi. > > From asyed at usc.edu Tue May 3 19:34:36 2005 From: asyed at usc.edu (Affan, Syed) Date: Tue May 3 19:37:25 2005 Subject: [Csci551-talk] Proj B Make files Message-ID: <000301c55051$d492f120$0402a8c0@A2D> Hello everyone, I know that a substantial number of people for project A got regraded b/c their make files did not compile on the grader cs551 account, but did so on their on account (as told to me by the grader). I am experiencing similar behavior this time. I would like all of such students to please email me their names so that I can quickly check if they work this time. I can then make you ppl come to my office and regrade on priority. Thank you. Best Regards, Affan, Syed. From ygandhi at usc.edu Tue May 3 19:59:36 2005 From: ygandhi at usc.edu (yash gandhi) Date: Tue May 3 20:01:23 2005 Subject: [Csci551-talk] Unique urls Message-ID: Hi, there was a comment in one of the classes that URLs are mostly unique, what would be an example of urls not being unique? comments!!! Yash From ygandhi at usc.edu Tue May 3 21:49:09 2005 From: ygandhi at usc.edu (yash gandhi) Date: Tue May 3 21:50:12 2005 Subject: [Csci551-talk] Unique urls Message-ID: Hi I am sorry, i mixed up. Professor pointed out that urls are "mostly unique" what could be an example where urls are not unique? comments!!! yash ----- Original Message ----- From: yash gandhi Date: Tuesday, May 3, 2005 7:59 pm Subject: [Csci551-talk] Unique urls > Hi, > there was a comment in one of the classes that URLs are mostly unique, > what would be an example of urls not being unique? > comments!!! > Yash > > From nikhilbh at usc.edu Wed May 4 21:10:34 2005 From: nikhilbh at usc.edu (nikhil bhatia) Date: Wed May 4 21:11:10 2005 Subject: [Csci551-talk] Routing in Wireless Message-ID: <4a8eeb321c7f.42793a4a@usc.edu> In highly dynamic wireless network which routing should perform the best Source/Distance Vector/Link State Routing ? Energy : All should perform equally well as there will be regular updates Comm : DV = DSR > LS (hign comm cost) However I am not too convinced about what I have said . -Nikhil Bhatia From srubin at flash.net Wed May 4 22:19:55 2005 From: srubin at flash.net (S. Rubin) Date: Wed May 4 22:20:51 2005 Subject: [Csci551-talk] RE: Routing in Wireless Message-ID: <20050505051955.35777.qmail@web81302.mail.yahoo.com> I was just studying this area, and my understanding is that we should not use Distance Vector or Link State due to too much overhead. They have to constantly advertise their routes, even if there are no changes. Link State also needs a lot of storage space. Since DSR does not use periodic routing advertisement messages, it results in reduced network bandwidth overhead, reduced power usage as hosts can "sleep" or be on "standby". Routes are dynamically determined based on cached information and on the results of a route discovery protocol. Sophia In highly dynamic wireless network which routing should perform the best Source/Distance Vector/Link State Routing ? Energy : All should perform equally well as there will be regular updates Comm : DV = DSR > LS (hign comm cost) However I am not too convinced about what I have said . -Nikhil Bhatia -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.isi.edu/pipermail/csci551-talk/attachments/20050504/62a044c7/attachment.html From asyed at usc.edu Wed May 4 22:28:16 2005 From: asyed at usc.edu (Affan, Syed) Date: Wed May 4 22:29:58 2005 Subject: [Csci551-talk] RE: Routing in Wireless In-Reply-To: <20050505051955.35777.qmail@web81302.mail.yahoo.com> Message-ID: <000b01c55133$41dd95f0$0400a8c0@A2D> I would go with Sophia, DSR (esp with intelligent caching) the best option for highly mobile networks. Best Regards, Affan, Syed. -----Original Message----- From: csci551-talk-bounces@mailman.isi.edu [mailto:csci551-talk-bounces@mailman.isi.edu] On Behalf Of S. Rubin Sent: Wednesday, May 04, 2005 10:20 PM To: csci551-talk@mailman.isi.edu Subject: [Csci551-talk] RE: Routing in Wireless I was just studying this area, and my understanding is that we should not use Distance Vector or Link State due to too much overhead. They have to constantly advertise their routes, even if there are no changes. Link State also needs a lot of storage space. Since DSR does not use periodic routing advertisement messages, it results in reduced network bandwidth overhead, reduced power usage as hosts can "sleep" or be on "standby". Routes are dynamically determined based on cached information and on the results of a route discovery protocol. Sophia In highly dynamic wireless network which routing should perform the best Source/Distance Vector/Link State Routing ? Energy : All should perform equally well as there will be regular updates Comm : DV = DSR > LS (hign comm cost) However I am not too convinced about what I have said . -Nikhil Bhatia -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.isi.edu/pipermail/csci551-talk/attachments/20050504/cd558b5c/attachment.html From nikhilbh at usc.edu Wed May 4 22:33:24 2005 From: nikhilbh at usc.edu (nikhil bhatia) Date: Wed May 4 22:33:46 2005 Subject: [Csci551-talk] Re :Routing in Wireless Message-ID: <1e52b76a21a87.42794db4@usc.edu> I understand what you have said, however my question was targeted at dynamic wireless networks. In this scenerio there have to be regular updates( even in DSR). Infact I beleive is it not the same comparison of (BGP:Overlay Networks) justification . May be DSR switches from SLEEP to ACTIVE modes is not required due the inherent environment. I am still not not convinced :) -Nikhil Bhatia Reply To :- I was just studying this area, and my understanding is that we should not use Distance Vector or Link State due to too much overhead. They have to constantly advertise their routes, even if there are no changes. Link State also needs a lot of storage space. Since DSR does not use periodic routing advertisement messages, it results in reduced network bandwidth overhead, reduced power usage as hosts can "sleep" or be on "standby". Routes are dynamically determined based on cached information and on the results of a route discovery protocol. Sophia From srubin at flash.net Wed May 4 22:41:42 2005 From: srubin at flash.net (S. Rubin) Date: Wed May 4 22:42:46 2005 Subject: [Csci551-talk] finger tables Message-ID: <20050505054142.84196.qmail@web81305.mail.yahoo.com> I am fuzzy on how we get the third column in the finger table (successors). I get the first column (identifiers), and the second column (intervals), but cannot follow through for the 3rd. I re-read sections 4.2 ane 4.3 of the Chord paper several times, but it still escapes me. Can someone demystify it, please. Sophia -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.isi.edu/pipermail/csci551-talk/attachments/20050504/c6486646/attachment.html From nsolanki at usc.edu Thu May 5 12:37:56 2005 From: nsolanki at usc.edu (nainesh solanki) Date: Thu May 5 12:39:30 2005 Subject: [Csci551-talk] Re: RE: Routing in Wireless Message-ID: <2fb4823f236e6.427a13a4@usc.edu> Hi, I believe there are DV algorithms specially designed for the mobile Ad-hoc environments such as the AODV which are also on-demand and compute routes on the fly. I read a research paper that compared the performance of AODV and DSR which as far as I remember showed that AODV infact performed better than DSR in more dynamically changing wireless networks. My point is that we do have optimizations with DV algorithms that do perform well in mobile ad-hoc networks(or MANETs as they are popularly called). Nainesh ----- Original Message ----- > > Message: 3 > Date: Wed, 4 May 2005 22:28:16 -0700 > From: "Affan, Syed" > Subject: RE: [Csci551-talk] RE: Routing in Wireless > To: "'S. Rubin'" , > Message-ID: <000b01c55133$41dd95f0$0400a8c0@A2D> > Content-Type: text/plain; charset="us-ascii" > > I would go with Sophia, DSR (esp with intelligent caching) the best > option for highly mobile networks. > > > > Best Regards, > > Affan, Syed. > > > > -----Original Message----- > From: csci551-talk-bounces@mailman.isi.edu > [csci551-talk-bounces@mailman.isi.edu] On Behalf Of S. Rubin > Sent: Wednesday, May 04, 2005 10:20 PM > To: csci551-talk@mailman.isi.edu > Subject: [Csci551-talk] RE: Routing in Wireless > > > > I was just studying this area, and my understanding is that we should > not use > > Distance Vector or Link State due to too much overhead. They have to > constantly > > advertise their routes, even if there are no changes. Link State also > needs a lot of > > storage space. Since DSR does not use periodic routing advertisement > messages, > > it results in reduced network bandwidth overhead, reduced power > usage as > hosts can > > "sleep" or be on "standby". Routes are dynamically determined > based on > cached > > information and on the results of a route discovery protocol. > > Sophia > > > > In highly dynamic wireless network which routing should perform > the best > Source/Distance Vector/Link State Routing ? > > Energy : All should perform equally well as there will be regular > updates > Comm : DV = DSR > LS (hign comm cost) > > However I am not too convinced about what I have said . > > -Nikhil Bhatia > > > -------------- next part -------------- > An HTML attachment was scrubbed... > URL: http://mailman.isi.edu/pipermail/csci551- > talk/attachments/20050504/cd558b5c/attachment-0001.html > > ------------------------------ > > Message: 4 > Date: Wed, 04 May 2005 22:33:24 -0700 > From: nikhil bhatia > Subject: [Csci551-talk] Re :Routing in Wireless > To: csci551-talk@mailman.isi.edu > Message-ID: <1e52b76a21a87.42794db4@usc.edu> > Content-Type: text/plain; charset=us-ascii > > I understand what you have said, however my question was targeted > at dynamic wireless networks. > In this scenerio there have to be regular updates( even in DSR). > Infact I beleive is it not the same comparison > of (BGP:Overlay Networks) justification . May be DSR switches from > SLEEP to ACTIVE modes is not required due > the inherent environment. > I am still not not convinced :) > -Nikhil Bhatia > > > Reply To :- > > I was just studying this area, and my understanding is that we > should not use > > Distance Vector or Link State due to too much overhead. They have > to constantly > > advertise their routes, even if there are no changes. Link State > also needs a lot of > > storage space. Since DSR does not use periodic routing > advertisement messages, > > it results in reduced network bandwidth overhead, reduced power > usage as hosts can > > "sleep" or be on "standby". Routes are dynamically determined > based on cached > > information and on the results of a route discovery protocol. > > Sophia > > > > > > ------------------------------ > > Message: 5 > Date: Wed, 4 May 2005 22:41:42 -0700 (PDT) > From: "S. Rubin" > Subject: [Csci551-talk] finger tables > To: csci551-talk@mailman.isi.edu > Message-ID: <20050505054142.84196.qmail@web81305.mail.yahoo.com> > Content-Type: text/plain; charset="us-ascii" > > I am fuzzy on how we get the third column in the finger table > (successors).I get the first column (identifiers), and the second > column (intervals), but > cannot follow through for the 3rd. I re-read sections 4.2 ane 4.3 > of the > Chord paper several times, but it still escapes me. Can someone > demystify it, please. > > Sophia > -------------- next part -------------- > An HTML attachment was scrubbed... > URL: http://mailman.isi.edu/pipermail/csci551- > talk/attachments/20050504/c6486646/attachment-0001.html > > ------------------------------ > > _______________________________________________ > Csci551-talk mailing list > Csci551-talk@mailman.isi.edu > http://mailman.isi.edu/mailman/listinfo/csci551-talk > > > End of Csci551-talk Digest, Vol 10, Issue 4 > ******************************************* > From rmchandr at usc.edu Thu May 5 16:22:37 2005 From: rmchandr at usc.edu (rashmi chandrasekhar) Date: Thu May 5 16:23:06 2005 Subject: [Csci551-talk] Freenet paper. Message-ID: Can anyone please clarify how exactly the search for CHK happens ? I believe that every node should know all the SSKs stored by all their neighbors and we choose the closest neighbor (whose key closely matches the SSK), in other words, we do hill climbing. This ensures we get to the node that has this SSK but this does not ensure that the same node has the contents. This node merely gives the CHK (in most cases atleast). Does the CHK come back to the node that initiated the request ? How will the initiator then look for the CHK ? -Rashmi. From johnh at ISI.EDU Thu May 5 16:42:07 2005 From: johnh at ISI.EDU (John Heidemann) Date: Thu May 5 16:43:07 2005 Subject: [Csci551-talk] GHT question In-Reply-To: Message-ID: <200505052342.j45Ng7PC013425@dash.isi.edu> Your comments are both sort of on-track, but I think you're mixing two things in GHT. There are two different kinds of replication in GHT: - perimeter replicas (discussed in section 4.3, the paper just calls these "replica nodes") - structured replicas (discussed in section 4.4, figure 7 calls these "mirror points") These are two different mechanisms with two different goals. Perimeter replicas are primarily for *reliability*---since the data has to circle the location, we make copies of it at each perimeter node as it does so. Thus if any of those nodes fail, the data is stored elsewhere. Structured replicas are for *load balancing*. The idea is that if it's too expensive (in terms of energy or storage or whatever) to store everything that would map to one location in one location, we can store it in several locations. Of course, if we do that, we now need to check all those locations when we do a search. With these *two concepts* in mind, I comment about the specific comments below. On Mon, 02 May 2005 23:32:12 -0800, rashmi chandrasekhar wrote: > >I think data is initially replicated. But later as queries for events are generated, all mirror nodes are queried, so, it is sort of like distributed. Below is what I could gather from the paper. This is true for structured replicas. (But not for permiter replicas.) >What they are saying is when a hash of a key maps to a geographic location, all nodes will try to send all data to that node at that location (even if it has to traverse the entire perimeter in doing so) for storage (and queries too). If there are many events happening with that key (suppose weather forecast for every minute) this is when the home node becomes a hot spot. To avoid this, the initial home node data is mirrored at suitable nodes as described in the paper. Now, when any node in the network detects a packet (or rather event) that hashes to the home node, instead of doing a perimeter forwarding and sending it to the home node all the way round the perimeter, the node just sends it to the nearest mirror. The nearest mirror then consumes this packet. This way load is well balanced. When later queries arise in the network, they have to be sent to all the mirrors however, because each of them could have stored different event packets forwarded to them from their neigh >bors. Yes, this is all true for structured replicas. >If I am wrong in my understanding of this concept, please let me know. > >-Rashmi. > > >----- Original Message ----- >From: lubaina kitabi >Date: Monday, May 2, 2005 2:35 pm >Subject: [Csci551-talk] GHT question > >> Hello, >> In the GHT paper, in the section that talks about structured >> replication , i was just wondering if replication is a misnomer. >> From what I could gather in that part was that structured >> replcation is used to create mirror nodes so that the home/root >> node does not become a hotspot. That would imply that for the same >> key value, the home node would have some data while the remaining >> data would be distributed among the "mirrors". Thus , the data is >> not being "replicated " just "distributed". >> Please correct me if I am wrong. Yes, this statement is true. Distribution (hashing?) might be a better term. Note that there is ALSO perimiter replication, which is real replication (duplication of data), to improve reliability. -John Heidemann From johnh at ISI.EDU Thu May 5 16:57:36 2005 From: johnh at ISI.EDU (John Heidemann) Date: Thu May 5 16:58:05 2005 Subject: [Csci551-talk] Unique urls In-Reply-To: Message-ID: <200505052357.j45Nva79013796@dash.isi.edu> On Tue, 03 May 2005 21:49:09 PDT, yash gandhi wrote: >Hi >I am sorry, i mixed up. >Professor pointed out that urls are "mostly unique" >what could be an example where urls are not unique? >comments!!! The example we takled about in class was: if you download the page at www.yahoo.com (or www.google.com or microsoft.com or sun.com or cnn.com or many similar big sites) are you guaranteed to go to the same sever? or to get the same web page? (The answers are definintely not, and not necessarily (but usually).) (It is outside the scope of class, but see the paper Armando Fox, Steven D. Gribble, Yatin Chawathe, Eric A. Brewer, and Paul Gauthier. Cluster-Based Scalable Network Services. In _Proceedings of the 16th Symposium on Operating Systems Principles_, pp. 78-91. St. Malo, France, ACM. October, 1997. . for a discussion of why you might get different web pages in response to queries at Google or elsewhere, and for why you might not care that the responses are different!) For that matter, one could argue that even MAC addresses are not GAURANTEED unique, since some network cards can change their MAC address in software. (Although I think MAC addresses are uniquely assigned, and forgery or maliciousness is outside the scope of a reasonable definition of "unique".) -John Heidemann From johnh at ISI.EDU Thu May 5 17:04:45 2005 From: johnh at ISI.EDU (John Heidemann) Date: Thu May 5 17:06:06 2005 Subject: [Csci551-talk] Routing in Wireless In-Reply-To: <4a8eeb321c7f.42793a4a@usc.edu> Message-ID: <200505060004.j4604jG5014051@dash.isi.edu> On Wed, 04 May 2005 21:10:34 PDT, nikhil bhatia wrote: >In highly dynamic wireless network which routing should perform the best >Source/Distance Vector/Link State Routing ? > >Energy : All should perform equally well as there will be regular updates >Comm : DV = DSR > LS (hign comm cost) > >However I am not too convinced about what I have said . > >-Nikhil Bhatia More important than the choice of DV vs. LS is the choice of a priori vs. on-demand. An a priori protocol propagates new routing information whenver there is a network change. An on-demand protocol computes new routing information whenever there is a need to route to a new location, or when an existing route in use breaks. -John Heidemann From johnh at ISI.EDU Thu May 5 17:15:13 2005 From: johnh at ISI.EDU (John Heidemann) Date: Thu May 5 17:16:10 2005 Subject: [Csci551-talk] finger tables In-Reply-To: <20050505054142.84196.qmail@web81305.mail.yahoo.com> Message-ID: <200505060015.j460FDPl014236@dash.isi.edu> On Wed, 04 May 2005 22:41:42 PDT, "S. Rubin" wrote: >I am fuzzy on how we get the third column in the finger table (successors). >I get the first column (identifiers), and the second column (intervals), but >cannot follow through for the 3rd. I re-read sections 4.2 ane 4.3 of the >Chord paper several times, but it still escapes me. Can someone >demystify it, please. > >Sophia > > What is labeled "succ." in the tables in Figure 3 corresponds to finger[k].node in Table 1. It is described in the 2nd paragraph of section 4.3. Basically, for each interval you need the node that represents that interval. >From Table 1, we know finger[k].start = n+2^{k-1} % (2^m) finger[k].interval = [finger[k].start, finger[k+1].start) and finger[k].node = first node >= finger[k].start so "start" is where the interval starts, "interval" is the range it contains, and "node" is the node that represents it. If there were nodes for all keys, then finger[k].node would always be finger[k].start, the first node in the interval. But since there are gaps, we let it "slip forward" from the start to the first node that actually exists. So for finger table for node 0 in figure 3b, finger[1].start = 1 // we start with the next node finger[1].interval = [1,2) // since we're finger[1] we cover only 2^0 = 1 key finger[1].node = 1 // since there is a node at 1 then finger[2].start = 2 // we start after finger[1].start finger[2].interval = [2,4) // at level 2 we cover 2^1 = 2 keys finger[2].node = 3 // since there is no node 2, we use node 3 and so on for finger[3], where node 0 actually represents that range since there are no other nodes in [4,0). Does this help? -John Heidemann From johnh at ISI.EDU Thu May 5 17:16:24 2005 From: johnh at ISI.EDU (John Heidemann) Date: Thu May 5 17:17:03 2005 Subject: [Csci551-talk] Freenet paper. In-Reply-To: Message-ID: <200505060016.j460GO3A014354@dash.isi.edu> On Thu, 05 May 2005 16:22:37 PDT, rashmi chandrasekhar wrote: > >Can anyone please clarify how exactly the search for CHK happens ? I believe that every node should know all the SSKs stored by all their neighbors and we choose the closest neighbor (whose key closely matches the SSK), in other words, we do hill climbing. This ensures we get to the node that has this SSK but this does not ensure that the same node has the contents. This node merely gives the CHK (in most cases atleast). > >Does the CHK come back to the node that initiated the request ? How will the initiator then look for the CHK ? One uses the same algorithm to search for ALL kinds of keys. CHK vs. SSK talks about how the keys are generated and used. They're all searched for using the hill-climbing algorithm. -John Heidemann From srubin at flash.net Thu May 5 18:36:38 2005 From: srubin at flash.net (S. Rubin) Date: Thu May 5 18:37:16 2005 Subject: [Csci551-talk] finger tables In-Reply-To: 6667 Message-ID: <20050506013638.7960.qmail@web81302.mail.yahoo.com> Got it, thank you! One last thing: In our slides on Chord, slide 10 "Key Distribution" states that data is distributed unevenly, but the Stoica paper on p.2, section 3, Load balance, states that the hash function is "spreading keys evenly over the nodes". Does their statement imply that the virtual nodes are used? In the last paragraph of section 4.2 they state that for simplicity, they dispense with the use of virtual nodes. Sophia John Heidemann wrote: On Wed, 04 May 2005 22:41:42 PDT, "S. Rubin" wrote: >I am fuzzy on how we get the third column in the finger table (successors). >I get the first column (identifiers), and the second column (intervals), but >cannot follow through for the 3rd. I re-read sections 4.2 ane 4.3 of the >Chord paper several times, but it still escapes me. Can someone >demystify it, please. > >Sophia > > What is labeled "succ." in the tables in Figure 3 corresponds to finger[k].node in Table 1. It is described in the 2nd paragraph of section 4.3. Basically, for each interval you need the node that represents that interval. >From Table 1, we know finger[k].start = n+2^{k-1} % (2^m) finger[k].interval = [finger[k].start, finger[k+1].start) and finger[k].node = first node >= finger[k].start so "start" is where the interval starts, "interval" is the range it contains, and "node" is the node that represents it. If there were nodes for all keys, then finger[k].node would always be finger[k].start, the first node in the interval. But since there are gaps, we let it "slip forward" from the start to the first node that actually exists. So for finger table for node 0 in figure 3b, finger[1].start = 1 // we start with the next node finger[1].interval = [1,2) // since we're finger[1] we cover only 2^0 = 1 key finger[1].node = 1 // since there is a node at 1 then finger[2].start = 2 // we start after finger[1].start finger[2].interval = [2,4) // at level 2 we cover 2^1 = 2 keys finger[2].node = 3 // since there is no node 2, we use node 3 and so on for finger[3], where node 0 actually represents that range since there are no other nodes in [4,0). Does this help? -John Heidemann -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.isi.edu/pipermail/csci551-talk/attachments/20050505/3c3a6848/attachment-0001.html From johnh at ISI.EDU Thu May 5 20:25:25 2005 From: johnh at ISI.EDU (John Heidemann) Date: Thu May 5 20:26:04 2005 Subject: [Csci551-talk] finger tables In-Reply-To: <20050506013638.7960.qmail@web81302.mail.yahoo.com> Message-ID: <200505060325.j463PPXM018554@dash.isi.edu> On Thu, 05 May 2005 18:36:38 PDT, "S. Rubin" wrote: >Got it, thank you! > >One last thing: >In our slides on Chord, slide 10 "Key Distribution" states that data is distributed >unevenly, but >the Stoica paper on p.2, section 3, Load balance, states that the hash function is >"spreading keys evenly over the nodes". Does their statement imply that the >virtual nodes are used? In the last paragraph of section 4.2 they state that for >simplicity, they dispense with the use of virtual nodes. If you look more closely at their results I think you can reconcile these statements. -John Heidemann From asyed at usc.edu Sun May 8 22:26:54 2005 From: asyed at usc.edu (Affan, Syed) Date: Sun May 8 22:29:15 2005 Subject: [Csci551-talk] Proj B grades Message-ID: <000001c55457$ba72c9a0$2e367d80@A2D> Hi all, Proj B grades have been handed out. You should have received an email by now. The grades consist of 2 parts: 60 pt for code test cases and another 60 for the writeup. I have rund the test cases and the email will have fairly detailed comments regarding your behavior. I will also upload the test cases and what is observed in each case. In view of the fact that a large number of students have code that doesn't compile, I will be holding office hours tommorow (Monday) from 6:00 pm to 9:00pm (only for code related regrades). We have to turn in the grades by Tuesday, so this is the only re-grade option for you. This mean that people who want to squabble over 4-5 marks will be discouraged. I will give higher preference to ppl who have gotten zero score. Hence I want there to be ****3 lines tommorow*****. 1) for people with zero marks 2) for ppl with 15-30 marks 2)all others. Kindly organize your self as a FIFO in these lines tommorow so that I can be as fair as possible to ppl who arrive early (now we might just get to use FQ over here ;) ). Secondly, since I have just 3 hrs, I will highly discourage squablling. I will first encourage all people to run the test cases on their own accounts, see if they observe what was required to get full credit. Only if it runs on their own, should they come for regrade. If not I will penalize an extra 10 marks from their grade. Also, if they find a minor change in the code will work, I will allow only 3 lines of code to change, with a penalty of -15 (so think if you really want to make that change!), and rerun the tests. Anything beyond that is not allowed. With a regarade, I will also look at the output more carefully and there is a chance of more marks being deducted as well. I will not give more than 5 min per person for a regrade request (other than for ppl with project compilation problems) Also please note that altought this might sound harsh, I have to implement such rules so that every one gets a fair chance at regrading. I have also tried to be fairly liberal in giving marks (some might not agree :)) , and the test cases are really simple - they donot test too much from your implementation. Hence I believe what I am doing will be fair. I am holding office hours at night to allow ppl having finals time to be there. Best Regards, Affan, Syed. From asyed at usc.edu Sun May 8 22:44:38 2005 From: asyed at usc.edu (Affan, Syed) Date: Sun May 8 22:46:16 2005 Subject: [Csci551-talk] Proj B grades In-Reply-To: <000001c55457$ba72c9a0$2e367d80@A2D> Message-ID: <000c01c5545a$33d90e60$2e367d80@A2D> Please wait a little while. The script that I wrote seems to have triggered a worm like danger signal from the smtp server :(. so I am redoing the sending. Also the first email is just with code grades, your write up and total grades will follow. Please be patient. If some one gets multiple my apologies. Thank you. Best Regards, Affan, Syed. -----Original Message----- From: csci551-talk-bounces@mailman.isi.edu [mailto:csci551-talk-bounces@mailman.isi.edu] On Behalf Of Affan, Syed Sent: Sunday, May 08, 2005 10:27 PM To: cs551@catarina.usc.edu Cc: csci551-talk@ISI.EDU Subject: [Csci551-talk] Proj B grades Hi all, Proj B grades have been handed out. You should have received an email by now. The grades consist of 2 parts: 60 pt for code test cases and another 60 for the writeup. I have rund the test cases and the email will have fairly detailed comments regarding your behavior. I will also upload the test cases and what is observed in each case. In view of the fact that a large number of students have code that doesn't compile, I will be holding office hours tommorow (Monday) from 6:00 pm to 9:00pm (only for code related regrades). We have to turn in the grades by Tuesday, so this is the only re-grade option for you. This mean that people who want to squabble over 4-5 marks will be discouraged. I will give higher preference to ppl who have gotten zero score. Hence I want there to be ****3 lines tommorow*****. 1) for people with zero marks 2) for ppl with 15-30 marks 2)all others. Kindly organize your self as a FIFO in these lines tommorow so that I can be as fair as possible to ppl who arrive early (now we might just get to use FQ over here ;) ). Secondly, since I have just 3 hrs, I will highly discourage squablling. I will first encourage all people to run the test cases on their own accounts, see if they observe what was required to get full credit. Only if it runs on their own, should they come for regrade. If not I will penalize an extra 10 marks from their grade. Also, if they find a minor change in the code will work, I will allow only 3 lines of code to change, with a penalty of -15 (so think if you really want to make that change!), and rerun the tests. Anything beyond that is not allowed. With a regarade, I will also look at the output more carefully and there is a chance of more marks being deducted as well. I will not give more than 5 min per person for a regrade request (other than for ppl with project compilation problems) Also please note that altought this might sound harsh, I have to implement such rules so that every one gets a fair chance at regrading. I have also tried to be fairly liberal in giving marks (some might not agree :)) , and the test cases are really simple - they donot test too much from your implementation. Hence I believe what I am doing will be fair. I am holding office hours at night to allow ppl having finals time to be there. Best Regards, Affan, Syed. From asyed at usc.edu Sun May 8 23:14:42 2005 From: asyed at usc.edu (Affan, Syed) Date: Sun May 8 23:16:10 2005 Subject: [Csci551-talk] ProjB grading policy and test cases Message-ID: <000d01c5545e$67308820$2e367d80@A2D> Hi all, The projb grading policy and test cases are now at TA web page: http://www-scf.usc.edu/~asyed/projb-policy.txt Please read carefulyl and run the test cases described in there with the expected results. Best Regards, Affan, Syed. From asyed at usc.edu Sun May 8 23:22:00 2005 From: asyed at usc.edu (Affan, Syed) Date: Sun May 8 23:24:10 2005 Subject: [Csci551-talk] Mistake in total score Message-ID: <000f01c5545f$6d7b4a70$2e367d80@A2D> Hi all, The ordering in your total score is incorect, please accept my apologies. The correct ordering is that: First line (which now says "Your Code score (max 60)") - is actually P6 write up score (max 30) Second Line (which now says "Phase 6 writup score (max 30)") - should be "Phase 7 writup score (max 30)" Third line (which now says "Phase 7 writeup score") - should be Code score (max 60) (which you should also get as a separate email). Last line is correct :). Thank you. Best Regards, Affan, Syed. From asyed at usc.edu Mon May 9 07:51:37 2005 From: asyed at usc.edu (Affan, Syed) Date: Mon May 9 07:53:16 2005 Subject: [Csci551-talk] Clarifications - PLEASE READ Message-ID: <002401c554a6$9c481080$0300a8c0@A2D> Hi all, Although I sent out the correction email about the grading order, not many ppl read that and I got a bunch of complains overnight saying their code score between the first and second email differed. The reason can be found in the email I sent out earlier... I mistakenly attached the wrong variable in the ordering of the msg :(. So you code score is the last before TOTAL. Again, my offcie hours will be 6:00 to 9:00 pm, with **3 seprate** queues. Kindly arrange your self accordingly. Finally, any writeup related question goes to Xi. Thank you. Best Regards, Affan, Syed. From asyed at usc.edu Mon May 9 08:04:23 2005 From: asyed at usc.edu (Affan, Syed) Date: Mon May 9 08:06:14 2005 Subject: [Csci551-talk] Any one who iddnt get scores Message-ID: <002901c554a8$67905bc0$0300a8c0@A2D> I am also hoping if any one didn't get his/her score - due to some email error or blocks that I couldn't figure - he should contact me asap. Also please tell all your frnds to check their score asap so their greviences can be redressed tonight. Best Regards, Affan, Syed. From xiw at usc.edu Mon May 9 09:10:35 2005 From: xiw at usc.edu (Xi Wang) Date: Mon May 9 09:14:30 2005 Subject: [Csci551-talk] office hours for writeup Message-ID: <427F8B7B.7030008@usc.edu> Dear All, I will be hold my normal office hours tomorrow (Tuesday 10:30am~12:30pm) for questions about writeup grading. This is for both classes. -Xi From asyed at usc.edu Mon May 9 14:52:42 2005 From: asyed at usc.edu (Affan, Syed) Date: Mon May 9 14:54:55 2005 Subject: [Csci551-talk] Email request for regrades Message-ID: <002f01c554e1$6e7e4170$cb357d80@A2D> CAUTION: This is for people who think their code works perfectly for some/all test cases(but not for those who might get zero or very low score, that might be an environment issue - those I would encourag coming to office hours only), you can send me email for regrades, that I can handle at night. However, and I plan to implement this strictly (this will also happen for ppl requesting such regrades at my ofice (other than ppl with environment/account problems), if during this regrade I see something other than what was initially wrong - I will deduct extra points. So just be careful (in fact I have already done that with one person). This will help reduce the queues, and hopefuly be fairer. Thankyou. Best Regards, Affan, Syed. From asyed at usc.edu Mon May 9 15:14:37 2005 From: asyed at usc.edu (Affan, Syed) Date: Mon May 9 15:16:46 2005 Subject: [Csci551-talk] DEN students Message-ID: <000001c554e4$813ccf90$37367d80@A2D> Since DEN students, who are long distance, have significantly difficulty in making it to the campus for exceptional regades (the zero compilation error problems).. I will take only those at my office phone (213) 740-4508. OR they can call me in uptil 4:30 pm at 310 448 8700 (only DEN/longdistance students). Thank you. Best Regards, Affan, Syed. From asyed at usc.edu Tue May 10 08:49:01 2005 From: asyed at usc.edu (Affan, Syed) Date: Tue May 10 08:52:03 2005 Subject: [Csci551-talk] No regular office hours today Message-ID: <000101c55577$caf17c60$0300a8c0@A2D> Hello all, I am canceling my regular TA office hours today, primarily because I am not at all feeling well today, and secondly b/c I put in a good > 5 hrs last night and was in the offcie till after 11:00 pm (there might be some correlation there :) ). Any way, I will try to handle any regrade request via email and let the concerned ppl know their results. Best Regards, Affan, Syed.