On Fri, Mar 14, 2003 at 11:15:16PM -0500, Michael Wiktowy wrote:
> I have been trying out the latest and greatest unstable builds. They 
> have been performing rather well showing very good resource usage ... 
> hovering around 40-50% load and being responsive. As of 
> freenet-unstable-20030313.tgz build the node has become catatonic after 
> 24 hours of uptime. It will not even load the gateway page.
> 
> I was using bandwidth limiting so that it could stay in the background 
> and my box would still be functional. If this functionality has been 
> removed inthe unstable builds also ... as seems to be indicated in 
> Matt's post here:
> http://hawk.freenetproject.org:8080/pipermail/devl/2003-March/004723.html 
> ...
> I would guess that that would be responsible for the recent DOS flood my 
> node is getting.
> 
> Removing this feature was like ripping off a bandage without the wound 
> having healed first it would seem.

You are misinterpreting. I disabled bandwidth limiting _FOR
AUTHORIZATIONS ONLY_. This is a very small amount of bandwidth, it
occurs once on every connection open. It was also essential for
performance since the bandwidth limiter seemed to be a major cause of
the ridiculously long connectingTime's we were seeing.
> 
> Some thoughts on healing the wound rather then building a bigger, better 
> bandage:
> 
> The load on the network placed by each message would be made up of a 
> number of factors:
> 1) the number of nodes that it has to pass through
>     - the more nodes that are involved with a particular message, the 
> more load each node will have to carry.
> 2) the time it spends transiting the network
>    - the more messages that are in flight in the network at a 
> particular moment in time the faster circular buffers will overflow, 
> timeouts will cancel pending messages which will have to be resent, etc.
> 3) the ratio of messages handled to messages generated for a particular node
>    - if there are more and more nodes that are leeches then the load 
> will increase for those that are active in the network and serving up 
> requests
> 
> Directions for improvement:
> 
> Item 1):
> More agressive path compression  ... as much as possible without 
> compomising anonymity
>    - maybe nodes should have the option of setting data source as the 
> node where it got the data from originally instead of itself even if it 
> was cached locally (improves anonymity, compresses path, makes sure that 
> ubernodes do not become the source of all freenety goodness ... doesn't 
> help prevent evil ubernodes that want to do this but it will help 
> unitentional ubernodeness)
> A guaranteed max and min path length that can be adjustible for tuning 
> purposes.
>    - I suppose that we have a guaranteed max (maxHTL) but there is no 
> limit of path compression as far as I know other then a fuzzy 
> equilibrium that will be established between caching and path 
> compression ... I am more convinced that it is a fuzzy doughnut rather 
> then a fuzzy ball ... another reason for the "original DataSource" 
> setting of nodes residing in the doughnut so that they can fill in the 
> center ... mmmmmm ... filled doughnuts ... I'm getting hungry.
> 
> Item 2):
> More utilization of already open persistent connections. Perhaps a bias 
> towards "close enough" still open connections rather then opening a new 
> one (and closing an old one) to a node that is slightly closer. Might 
> lead to a more statically linked mesh of nodes. Still needs a bit of pot 
> stirring to defy traffic analysis but perhaps not the tempest that the 
> routing table is now.

No way. We cannot compromize routing to reduce hop times, it will be a
disaster. What we can do is implement multiplexing to reduce the need
for multiple simultaenous connections to the same node, and nio to let
us keep 2-3x rtMaxNodes connections open at once without using
gazillions of threads.
> 
> Making the comm/encryption setup and tare down as quick as possible 
> should be a constant endeavour.

That is precisely what I was engaged in in the above mentioned change.
> 
> Item 3):
> The current trend towards making nodes more permenant will help this a 
> lot. There are still nodes that might like to be permanent but cannot 
> for different reasons
>    - behind NAT - perhaps some sort of polling solution or a non-NATed 
> buddy node system (sort of like a C/O post office addresss) could help 
> integrate some of these inaccessible nodes into the network
>    - local load too high to run on a desktop with other things going on 
> - if Freenet bogs down your desktop then people will not keep it on and 
> you will get a lot of "transient" permanent nodes screwing with the 
> routing tables
>     - nodes running for only a short period of time - if a node can pop 
> out without disruption (connection handoff/no new connections while 
> others finish off while shutting down) and become a functional part of 
> the network quickly upon start up (either first time or after a 
> temporary downtime) freenet will be a better place.

Routing has become rather more dynamic recently.
> 
> OK ... enough rambling ... I hope someone has read this far and 
> understood what I am talking about but I doubt it.
> 
> /Mike

-- 
Matthew Toseland
toad at amphibian.dyndns.org/amphibian at users.sourceforge.net
Full time freenet hacker.
http://freenetproject.org/
Freenet Distribution Node (temporary) at 
http://80-192-4-36.cable.ubr09.na.blueyonder.co.uk:8889/0bIANZmR~K8/
ICTHUS.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: 
<https://emu.freenetproject.org/pipermail/devl/attachments/20030315/f8c2fc62/attachment.pgp>

Reply via email to