Re: [freenet-support] maxHopsToLive in Freenet configuration file
[EMAIL PROTECTED] wrote: maybe the default values for rtMaxNodes=50 rtMaxRefs=50 are too low, so your node's connectivity is bad? i increased them both to hold up to 500 entries and within an hour i got 174 noderefs! (okay, i did not have a permanent node, so they had just 1-25 keys assigned, but what the heck ;) Oh my gosh, you are trying to hold 500x500=25 keys! I am quite sure this would end up with a big use of store space, and unless you really have a big one I believe it would eat up space from actual content `:| i wonder if there is any good reason why these values are limited to 50 entries and not to +INF ? I am quite sure about the MaxNodes: if you have a perfect routing table (network is fully connected: everyone knows everyone) then you will probably get things with HTL=1 or something very close. On answer, you and you alone (maybe one or two more) will have cached the content. Hence, things will get much worse (if a couple of nodes go down you loose the files). Also, bigger tables mean more CPU consumption. Also, results get cached in paths. If you get lots of alien keys (those that are not near your node's assigned key) you will loose important data (the one you should be having), causing great havock. Data key localisation is one of the things that make freenet routing algo work. Maybe there are better reasons, or mine has a flaw in it. Anyway, that's what I think :) -- --- TLD Oh, how uncomfortable that word must feel on your lips: evil. Good...there is no good, there is no evil. There is only flesh, and the patterns to which we submit it. [Pinhead] ___ support mailing list [EMAIL PROTECTED] http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/support
[freenet-support] maxHopsToLive in Freenet configuration file
Hi Free Netters, I just joined this list, so if the issue has been discussed before please point me to the right location. Search of the archive did not give me the required results. I tend to believe that information in freenet is available which I can not find reliably. While I can not prove that, the behaviour in extracting information seems to indicate it. Sometimes information drops in after repeating requests for days. I am running a permanent node with a large data- store, so I should have good prerequisites regarding rou- ting. What I found in the freenet configuration file is the entry maxHopsToLive, which seems to have a default value of 25 and is described to limit the HTL value of requests passing a node. If this limit is active it would mean that using a HTL value beyond that limit will be of no help, since most likely the majority of users do not change that default and the HTL will be limited at the first node the request hits. Is that assumption correct? I do not really understand why this value is introduced and used. There is the risk that information available can not be found, right? And I understand that circular requests are terminated anyhow, right? Please enlighten me, whether my assumptions are comple- tely wrong, whether they are right but default is enough to search a whole universe full of freenet nodes or whe- ther the default should be changed. Are there other means to increase the probability to find information in freenet? Best regards, Thomas ___ support mailing list [EMAIL PROTECTED] http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/support
Re: [freenet-support] maxHopsToLive in Freenet configuration file
On Mon, Dec 09, 2002 at 11:56:30PM +0100, Thomas Wuensche wrote: Hi Free Netters, I just joined this list, so if the issue has been discussed before please point me to the right location. Search of the archive did not give me the required results. I tend to believe that information in freenet is available which I can not find reliably. While I can not prove that, the behaviour in extracting information seems to indicate it. Sometimes information drops in after repeating requests for days. I am running a permanent node with a large data- store, so I should have good prerequisites regarding rou- ting. What I found in the freenet configuration file is the entry maxHopsToLive, which seems to have a default value of 25 and is described to limit the HTL value of requests passing a node. If this limit is active it would mean that using a HTL value beyond that limit will be of no help, since most likely the majority of users do not change that default and the HTL will be limited at the first node the request hits. Is that assumption correct? Yes. I do not really understand why this value is introduced and used. There is the risk that information available can not be found, right? And I understand that circular requests are terminated anyhow, right? Because we want to be able to limit the length of time and the amount of network resources that a request takes. A request only goes to a certain number of nodes, the hops-to-live. This means that unlike gnutella, freenet should be able to scale as a single contiguous network, and an attacker who just sends zillions of requests only gets (max HTL) * (number of requests) load/bandwidth for his money, rather than (max HTL) * (number of nodes). Please enlighten me, whether my assumptions are comple- tely wrong, whether they are right but default is enough to search a whole universe full of freenet nodes or whe- ther the default should be changed. Are there other means to increase the probability to find information in freenet? We may want to increase the default maximum HTL in 0.5.1, however there are lots of nodes left over from 0.5.0 and its immediate successors, which would tend to prevent this, short of forking the network... Best regards, Thomas -- Matthew Toseland [EMAIL PROTECTED] [EMAIL PROTECTED] Freenet/Coldstore open source hacker. Employed full time by Freenet Project Inc. from 11/9/02 to 11/1/03 http://freenetproject.org/ msg02323/pgp0.pgp Description: PGP signature
Re: [freenet-support] maxHopsToLive in Freenet configuration file
Matthew, Thanks for your fast and precise answer. I have removed all text where I think it does not need more discussion. The following points remain: Snip I do not really understand why this value is introduced and used. There is the risk that information available can not be found, right? And I understand that circular requests are terminated anyhow, right? Because we want to be able to limit the length of time and the amount of network resources that a request takes. A request only goes to a certain number of nodes, the hops-to-live. This means that unlike gnutella, freenet should be able to scale as a single contiguous network, and an attacker who just sends zillions of requests only gets (max HTL) * (number of requests) load/bandwidth for his money, rather than (max HTL) * (number of nodes). Oh, agreed, protecting freenet against attackers definitely is worth that considerations. However with a growing net the percentual reach of a search is getting smaller. A net which is willing to give only a fraction of the information it holds to somebody searching for the information of course is also of limited value. The problem specially is that information that is new or rarely requested does propagate badly. From the view of a node searching for the information it may become available only after somebody in between has also requested it. The only chance besides that might be a shifted search horizon of the requester. Is there a chance that my node gets to know more nodes out there in the freenet universe and to ask them in an arbitrary sequence? Or is there another chance to find infor- mation which is beyond my search range (as seen at a single point in time)? This would at least help me to find the infor- mation, even if it costs me more bandwidth. Repeatedly asking the nodes that did not know it before in the hope they might know it in the future will also significantly increase network load, but with less probability of success for the regular user. Please enlighten me, whether my assumptions are comple- tely wrong, whether they are right but default is enough to search a whole universe full of freenet nodes or whe- ther the default should be changed. Are there other means to increase the probability to find information in freenet? We may want to increase the default maximum HTL in 0.5.1, however there are lots of nodes left over from 0.5.0 and its immediate successors, which would tend to prevent this, short of forking the network... This would be welcome, please increase it to a value that holds some margin for future growth. The impact of infor- mation that is not delivered may be bigger than the impact of an overload attack. The impact of an overload attack might also be limited by limiting the bandwidth a node is willing to provide to a single requester, would that be reasonable? With regard to new versions, wouldn't it help if a certain percentage of the nodes had higher limits? These would pro- vide paths of deeper information flow into the network, even if other paths stay shallow. And of course it could easily be improved without a version change, if people would change it manually. Is there a promi- nent place where that could be published? Best regards, Thomas ___ support mailing list [EMAIL PROTECTED] http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/support
Re: [freenet-support] maxHopsToLive in Freenet configuration file
On Tue, Dec 10, 2002 at 03:06:39AM +0100, Thomas Wuensche wrote: Matthew, Thanks for your fast and precise answer. I have removed all text where I think it does not need more discussion. The following points remain: Snip I do not really understand why this value is introduced and used. There is the risk that information available can not be found, right? And I understand that circular requests are terminated anyhow, right? Because we want to be able to limit the length of time and the amount of network resources that a request takes. A request only goes to a certain number of nodes, the hops-to-live. This means that unlike gnutella, freenet should be able to scale as a single contiguous network, and an attacker who just sends zillions of requests only gets (max HTL) * (number of requests) load/bandwidth for his money, rather than (max HTL) * (number of nodes). Oh, agreed, protecting freenet against attackers definitely is worth that considerations. However with a growing net the percentual reach of a search is getting smaller. A net which Yes, it sounds bad if you don't understand how it works. Read some of the papers on freenetproject.org. Basically, the idea is that with each hop the network routes the request closer to the location of the data. This is accomplished by freenet routing, and is explained in depth on the web site, but the gist of it is that different nodes specialize in different sorts of keys, and nodes keep routing tables, which tell them which nodes have successfully sent back data for which kinds of keys. is willing to give only a fraction of the information it holds to somebody searching for the information of course is also of limited value. The problem specially is that information that is new or rarely requested does propagate badly. From the view Yeah. The network has some problems at the moment; I do suspect that we will need to increase the maximum HTL, but that's probably not the main issue causing slowness and unreliability at the moment on freenet. And it's better than it has been for a long time. of a node searching for the information it may become available only after somebody in between has also requested it. The only chance besides that might be a shifted search horizon of the requester. Is there a chance that my node gets to know more nodes out there in the freenet universe and to ask them in an arbitrary sequence? Or is there another chance to find infor- mation which is beyond my search range (as seen at a single point in time)? This would at least help me to find the infor- mation, even if it costs me more bandwidth. Repeatedly asking the nodes that did not know it before in the hope they might know it in the future will also significantly increase network load, but with less probability of success for the regular user. Please enlighten me, whether my assumptions are comple- tely wrong, whether they are right but default is enough to search a whole universe full of freenet nodes or whe- ther the default should be changed. Are there other means to increase the probability to find information in freenet? We may want to increase the default maximum HTL in 0.5.1, however there are lots of nodes left over from 0.5.0 and its immediate successors, which would tend to prevent this, short of forking the network... This would be welcome, please increase it to a value that holds some margin for future growth. The impact of infor- mation that is not delivered may be bigger than the impact of an overload attack. The impact of an overload attack might If freenet routing works, then a fairly low HTL should be sufficient for a very large network - again, there is more detail on the website. also be limited by limiting the bandwidth a node is willing to provide to a single requester, would that be reasonable? Freenet without routing is nothing. We must get freenet routing to work better. We are working on this. We will probably increase the default max HTL at some point, but there is a lot we can do other than that to improve performance. With regard to new versions, wouldn't it help if a certain percentage of the nodes had higher limits? These would pro- vide paths of deeper information flow into the network, even if other paths stay shallow. No. A node receives a request at hops-to-live 24, it finds a node that can accept the request and forwards it to that node at hops-to-live 23, that node then finds another node and forwards it at 22 etc etc. And of course it could easily be improved without a version change, if people would change it manually. Is there a promi- nent place where that could be published? Yuck. We would never get all freenet users to update it, and besides, we want to make freenet as low maintenance as reasonably possible. Best regards, Thomas -- Matthew Toseland [EMAIL PROTECTED] [EMAIL PROTECTED] Freenet/Coldstore open source hacker. Employed full time by Freenet Project
