> From: Ian Clarke <ian at freenetproject.org>
> Date: 28 Oct 2002 18:23:33 -0800
> 
> It seems that the flood of new users over the past day has caused 
> Freenet some growing pains - it wouldn't surprise me if the network 
> grew 10X or even 20X over the past 15 hours.
> 
> Unfortunately this rapid growth has created a storm of announcement 
> messages which appear to be flooding the network, I count 6070 
> NodeAnnouncements going through my node today, and only 3101 
> AnnouncementReplies.  This means that half of the nodes which 
> announced through me didn't get a response, and will therefore be 
> trying their Announcement again later - given this the overloading is 
> not really that surprising.  By comparison there were only 1401 
> DataRequests, and 83 DataReplies!  Presumably the NodeAnnouncements
> 
> Toad and I had a discussion by phone, and agreed that we should 
> release 0.5.0.1 ASAP which incorporates Oskar's new load balancing 
> code, which may or may not alleviate the problem.
> 
> Some may ask why we are releasing this code when it is relatively 
> untested, but the problem is that the only way to test it is wide 
> deployment.
> 
> We must also think about how to modify the announcement behavior so 
> that this doesn't happen the next time Freenet hits the papers 
> (something we can sometimes provoke, but often have no control over).
> 
> Ian.

Never discount the possiblity of sabotage. I hate to sound paranoid but
the load on Freenet came on a little too high a little too fast to be a
slashdotting. For regular websites getting slashdotted, they get a quick
flood because everyone is *already* browsing and it is simple as
clicking a small link to place load on their servers. With Freenet it is
an order of magnitude more difficult. They have to download the freenet
software, install Freenet, debug their java JREs, undoubtedly download
and install a new JRE, install Freenet again, then figure how to get to
the gateway page via a browser. This is not something that would
simultaneously happen for many people exactly at the same time that the
story came out.

Take a look at the webpage logs to see how many more visitors it got
that day. I would bet that it is not as many as you think. Why do I
think that? Where is the enevitable flood of whining "Freenet sucks ...
I couldn't get anything through it/ I couldn't get it to run/ etc." in
the story response or in the support mailing list. There are a handful.
Nothing compared to what I'd have expected for 10-20x the size of
existing Freenet (currently estimated fromthe number of unique IPs shown
in the routing table at about ~4000 nodes) trying to get things to run
all at once.

How could someone sabotage Freenet? All that I can say is from the
symptoms that I saw happening on my node. My thread count was pegged to
the maximum and then some and the number of Incoming Requests and
Incoming Inserts dropped down to zero ... not near zero but *zero*. If
someone found a trivial way to make a Freenet node generate a thread and
then hold it open for a timeout period then only a few machines could
tie up all the seednodes for as long as they wanted. What would happen
if a connection to a node was started but then dropped on the floor?
What would happen if a bunch of fake ennouncements were spammed to all
known nodes?

I'm not saying that the node is bullet-proof and that this wasn't
actually just a whole lot of new nodes joining. But there doesn't seem
to be the strong indications of disappointment that would accompany the
utter failure of Freenet for a mob of new users. It makes me suspicious.
Just don't go off all half-cocked trying to treat the symptoms that were
shown during this stress test.

Mike


_______________________________________________
devl mailing list
devl at freenetproject.org
http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to