Update of /cvsroot/freenet/freenet/src/freenet/node
In directory sc8-pr-cvs1:/tmp/cvs-serv28319/src/freenet/node
Modified Files:
SmartFailureTable.java FailureTable.java
Log Message:
have fun toad :)
Index: SmartFailureTable.java
==
Update of /cvsroot/freenet/freenet/src/freenet/node
In directory sc8-pr-cvs1:/tmp/cvs-serv25916/src/freenet/node
Modified Files:
SmartFailureTable.java FailureTable.java
Log Message:
initial commit
Index: SmartFailureTable.java
Update of /cvsroot/freenet/freenet/src/freenet/node
In directory sc8-pr-cvs1:/tmp/cvs-serv23686/src/freenet/node
Added Files:
SmartFailureTable.java
Log Message:
initial commit
--- NEW FILE: SmartFailureTable.java ---
package freenet.node;
import freenet.Core;
import freenet.Key;
import
Can anyone familiar with xsl list the differences between filtering html
for anonimity-compromising content and filtering xsl transformations?
If an xml file is filtered agains the current rules (no off-freenet
links, no actions) and its corresponding xsl is made sure not to contain
any such t
lets not forget this other favorite...
Oct 22, 2003 10:32:31 PM (freenet.node.states.request.TransferReply,
QThread-260, ERROR): Send failed, cache broke
n! for freenet.node.states.request.TransferReply:
key=92b8d318833e6d7e5e8a8d18388dbbe9ba94df00140302, hopsToLive=1
3, id=bfac2582bdef4813, [EM
To all fans of this eternal error message, here's some more: (build 6273)
Oct 22, 2003 9:55:17 PM (freenet.node.Node, QThread-174, ERROR): Error
while receiving message freenet.Message: Ac
cepted @[EMAIL PROTECTED] for tcp/connection:
55930>209.210.55.3:11902,freenet.transport.tcpConnec
[EMAIL
java.lang.NullPointerException
at
freenet.interfaces.BaseLocalNIOInterface.intAddress(BaseLocalNIOInterface.java:37)
at
freenet.interfaces.BaseLocalNIOInterface.hostAllowed(BaseLocalNIOInterface.java:189)
at
freenet.interfaces.BaseLocalNIOInterface.dispatch(BaseLocalNIOInter
The state chain code uses the reflection api heavily. Official
tutorials from Sun recommend reflection to be used only as a last
resort, i.e. if the problem cannot be solved any other way. RTTI in
win32 adds tons of overhead.
Is anyone out there 100% positive that reflection in java is chea
10/12/03 9:49:00 PM EDT 2704.2711934156378 21732.06106603715 1.0
288242.0 2430
10/12/03 9:50:00 PM EDT 6820.178354978355 42211.21986746211 1.0
649655.0 2310
10/12/03 9:51:00 PM EDT 4288.004818221638 24078.73450596664 1.0
367618.0 2283
10/12/03 9:52:00 PM EDT 183713.43742735373 271629.6005761425
sorry for the html, images were removed.
OCM lists some connections as incoming at port -2, -1...
-1
208.39.42.23:35634 0 2,493 Bytes - None0 s 1 s
35713
207.32.9.50:11975 366 299 KiB - 294 KiB 0 s
6:46
-2
207.32.
Since Freesites are placed on the gateway page to allow people to find
freesites more easily, we should probably drop YoYo now that he has
deliberately crippled his site to promote a network split.
Shame.
Ian.
Or emply the slightly less strategy of ignoring them until they go away or the slight
Some very ill-informed people are spreading the rumours to downgrade to
random builds without even knowing what is in those builds and without
making the effort to read through the changelog and make an informed
decision. If you're going to downgrade, at least stay away from
anything pre-5008
because they will be separate networks with different usage patterns.
as long as people are running frost and fuqid you'll see 99% identical patterns. Those two clients amount for most of the traffic anyways (and no this is not a plug for frost because fuqid generates far more traffic).
Note that
I found and managed to run a snapshot of TFEE from June 29th. It needed
slight tweaking to compile but after that everything went smooth
except... speed.
It needs a lot more than my athlon 1.2 with 512MB ram can provide. Is
there anybody on the list who would wish to run TFEE? The requiremen
I had a ~5 minute 100% cpu spike with latest cvs (which includes
lostlogic's blacklist patches). A stack dump gives tons of threads
(including wsl) waiting to lock a lru in ocm , and the thread holding it
waiting to lock a hashtable in ngr.references(). Nothing visibly holds
that monitor.
Currently 3 out of 5 links on the gateway are dead, with one of them not
having had an update for almost 6 months. TFE hasn't been updated in
almost a month as well. Now, if CofE was not interested in maintaining
his flog he would at least have the courtesy to inform the project. He
knows it
AbstractSelectorLoop register/unregister should be calling sel.wakeup().
No. sel.wakeup() will not release the locks on the sets and will cause registering the keys to block until the selector thread is run again.
AbstractSelectorLoop processWaiters that first while loop looks bad.
I don't see ho
Update of /cvsroot/freenet/freenet/src/freenet/node
In directory sc8-pr-cvs1:/tmp/cvs-serv25642/src/freenet/node
Modified Files:
ConnectionOpener.java
Log Message:
accounting fix
Index: ConnectionOpener.java
===
RCS file: /c
Update of /cvsroot/freenet/freenet/src/freenet
In directory sc8-pr-cvs1:/tmp/cvs-serv1889/src/freenet
Modified Files:
ConnectionHandler.java
Log Message:
notify everybody on the sendingQueue in terminate()
Index: ConnectionHandler.java
Update of /cvsroot/freenet/freenet/src/freenet
In directory sc8-pr-cvs1:/tmp/cvs-serv3206/src/freenet
Modified Files:
ConnectionHandler.java
Log Message:
fix braino, some questions
Index: ConnectionHandler.java
===
RCS file:
Update of /cvsroot/freenet/freenet/src/freenet
In directory sc8-pr-cvs1:/tmp/cvs-serv3866/src/freenet
Modified Files:
ConnectionHandler.java
Log Message:
fixed a rare fcp fec bug
Index: ConnectionHandler.java
===
RCS file: /
Update of /cvsroot/freenet/freenet/src/freenet
In directory sc8-pr-cvs1:/tmp/cvs-serv23612/src/freenet
Modified Files:
OpenConnectionManager.java
Log Message:
oops
Index: OpenConnectionManager.java
===
RCS file: /cvsroot/fre
Update of /cvsroot/freenet/freenet/src/freenet
In directory sc8-pr-cvs1:/tmp/cvs-serv32550/src/freenet
Modified Files:
OpenConnectionManager.java ConnectionHandler.java
Log Message:
some tweaks, speedups, loggings...
Index: OpenConnectionManager.java
=
Update of /cvsroot/freenet/freenet/src/freenet
In directory sc8-pr-cvs1:/tmp/cvs-serv26819/src/freenet
Modified Files:
ConnectionHandler.java
Log Message:
if logDebug...
Index: ConnectionHandler.java
===
RCS file: /cvsroot/f
Seems like the high cpu usage is due mainly to heavy churn of new
connections being opened on popular nodes. Eventually a feedback loop
occurs: new connections are being opened to reject queries due to bw
limits, and those negotiations in turn take up the bandwith causing more
QRs.
My node re
Update of /cvsroot/freenet/freenet/src/freenet/node
In directory sc8-pr-cvs1:/tmp/cvs-serv6510/src/freenet/node
Modified Files:
Main.java ConnectionOpener.java
Log Message:
new diag
Index: Main.java
===
RCS file: /cvsroot/fr
Update of /cvsroot/freenet/freenet/src/freenet/transport
In directory sc8-pr-cvs1:/tmp/cvs-serv26857/src/freenet/transport
Modified Files:
AbstractSelectorLoop.java
Log Message:
ignore cancelled key excptions
Index: AbstractSelectorLoop.java
==
Update of /cvsroot/freenet/freenet/src/freenet/node
In directory sc8-pr-cvs1:/tmp/cvs-serv2254/src/freenet/node
Modified Files:
Main.java
Log Message:
added new diag hoping to catch loops
Index: Main.java
===
RCS file: /cvsr
Update of /cvsroot/freenet/freenet/src/freenet/transport
In directory sc8-pr-cvs1:/tmp/cvs-serv2254/src/freenet/transport
Modified Files:
AbstractSelectorLoop.java
Log Message:
added new diag hoping to catch loops
Index: AbstractSelectorLoop.java
=
A. If both are guaranteed to exist then A is better.
at least 90% of the time neither file will exist.
If both maxout without finding anything, then you ask the same number of nodes
ether way.
In theory yes. In practice, there will be plenty of QR-ing, restarting, etc. If one request RNFs, t
Suppose I insert two files with the same size but different CHKs at the
same htl. Both files will be equally popular (this can be guaranteed).
I need to get just one of the files and want to do that in a way that
will be least stressfull to the network. Which one is the optimal solution:
a)
Update of /cvsroot/freenet/freenet/src/freenet/transport
In directory sc8-pr-cvs1:/tmp/cvs-serv15768/src/freenet/transport
Modified Files:
Tag: ngrouting
AbstractSelectorLoop.java
Log Message:
reworked closing
Index: AbstractSelectorLoop.java
===
Update of /cvsroot/freenet/freenet/src/freenet/transport
In directory sc8-pr-cvs1:/tmp/cvs-serv14994/src/freenet/transport
Modified Files:
AbstractSelectorLoop.java
Log Message:
reworked closing
Index: AbstractSelectorLoop.java
We're having some issues with locking which are most apparent on
nptl-enabled systems. Our usual stance on the subject has been that
nptl is screwed; the corporate articles from $un and redhat state that
the synchronization mechanism has been reworked to be more efficient.
I'm inclined to buy
Or maybe it
is only so that my node has tried to Terminate the *same* (1c17c3a)
ConnectionHandler 16k times during those seconds?
more probably the same connectionHandler gets enqueued over and over to
the closeQueue. I believe I removed closeUniqueness some time ago while
debugging .. somet
I am still investigating the very long messageSendTime's
On a very busy node (40-60kqph) I've yet to get those notorious long messageSendTimes, both with 2.6 and 2.4.20
8/13/03 3:00:00 PM EDT 91.80413460973747 278.48429234684437 0.0 5570.0
51081
8/13/03 4:00:00 PM EDT 95.3804948
I don't usually get involved in eternal flamewars even when I have
opinion and facts on the subject. So, here are just numbers:
maximumThreads = 500
8/5/03 10:00:00 PM EDT871187070.9995408104695213
8/5/03 11:00:00 PM EDT748773690.9842393482035529
8/6/03 12:00:00 AM EDT
Currently we are seeing very high times after the route is found until
the message leaves the node. They are in the order of 30 seconds and in
some cases more than 2 minutes. We can chase the reasons everywhere,
but the basic truth is that since we RoundRobin to all sending
connections, the m
I don't remember if the current message headers have a byte which
designates which session they belong to. If they do, we should not
worry about parsing messages that do not have trailing fields: on the
sending side we make sure we send them atomically, and on the receiving
side we just parse
the socket leak manifests itself only on heavily overloaded nodes. My
node's uptime so far has been 19 hours, but its load never got above
10kqph, and the number of fds hovers around 600. During the /.-ing, my
load was rarely below 20kqph, and the node would use up to 3000 fds in 5
hours.
S
The bottom line is that if these issues persist and can't be worked
around then we should back-merge to pre-NIO and continue from there
with NGrouting.
As you can see the reports are contradicting, and I bet that at least two thirds of those who report freezing have not read the README file and hav
The first SVM will predict if the request will succeed or fail.
that's going one step ahead of what we thought so far.. interesting idea. However, i'm against using it on nodes for which we have
very little learning data. When a new node is introduced to the network, its is very inefficient at f
hello kaffe zealots,
now there's a branch in cvs called "non-nio". Everything non-nio
dependend will be backported from branch stable BY YOU!!! Please keep
the build numbers the same as the stable build you backported from.
I'm not sure whether we'll provide binaries for download, its up to
libsvm is really cool, even has a idi^H^H^H beginner's guide to using it ;)
do you think we could use that to make global estimates across all
nodes, i.e. instead of estimating the response time for each node in the
rt and then choosing the best, could we add the routingTable contents as
a disc
FFT gets very expensive when the number of reference points grows too much, but between 5 to 50 reference points its doable on >current cpus. (not sure about those numbers, could be a lot more/less)
did my homework. Its N*log(N), and is significantly faster than what I thought - the fortran lib t
In that are you assuming that there will only be one area of
specialization? - because we can't assume that.
No, he said it passes through all the points.
There are two situations where I believe the benefits of interpolation will outweight the cpu cost.
First one is of new to young nodes who h
Take a look at The Tower's flog, he seems to have found a way to
automatically link to the latest edition of an edition-based freesite.
[EMAIL PROTECTED]/TFEE//flog.html
___
devl mailing list
[EMAIL PROTECTED]
http://hawk.freenetproject.org:8080/cgi-bi
Ok, some of you may not like this, but if a windows user is advanced
enough to know what the settings on the advanced and geek tabs do, they
should also be advanced enough to edit the .ini files themselves. The
config tool as it is quite intimidating in its complexity to new users.
The reacti
* We need a diagnostic which tracks the difference between estimated and
actual routing times so that we can guage how well NG is working
Could that diagnostic please be chosen in a way that it would not compromise anonymity?
This way it could be reported globally (like we report load now).
We c
Hi,
if you plan on publishing a paper or something on NGRouting it may be a
good idea to investigate further into what kind of behavior does the
graph keyspace vs. time evolves. I'm very interested in studying which
method of estimating response time is best mathematically because it may
unve
Well, I once tried to make a liner time function to derive the formula using
an Nth degree polynomial. At the time is was just an exercise, so I gave up
after the math became too much of a bitch. (I think I could do it.)
??? For N sample points you get Nth degree polynomial by creating NxN matrix
Possible improvements over using a fixed number of reference points and
straight lines between them would be :
I. Non-linear estimators: instead of straight lines, user some sort of
interpolation to create the graph :
a. If the number of points is fixed around 10, we could use newton
interpola
I suggest we do not throttle incoming bandwith at all.
The main reason is that we do not throttle messages with trailing fields
differently from the ones that don't. For the end-user that means that
they cannot set separate limits for routing data as opposed to file
transfer. Most end users i
Due to recent morbid discoveries, we need to update the list of working
combinations for freenet.
So far:
Redhat 9 + 1.4.2 client & server : DOUBLEPLUSFSCKDEDUP
Redhat 9 + 1.4.1 server : a little better
Redhat 9 + 1.4.1 client : probably a little better than server, not
tested extensively
Non
>The only ones that support NIO, to my knowledge, are gcj and Classpath.
there is a NBIO library (I forgot the url) which is fully supported by
kaffe and gcj, and is compatible with 1.3 and has seen heavy production use.
So if you want to help freenet run on non-proprietary non-1.4 platforms,
y
The FCP specification is not directly reachable from the website. The
old (twiki & tiki) sites had it, even translated as far as I remember.
Could someone please put a link to it somewhere? (probably in the tools
section).
The only relevant document I could find is an year and a half old an
My node deadlocked after 15 hrs of uptime in a
particularily-hard-to-debug deadlock. We are also still leaking memory,
only very slowly. To debug these issues we'll need a _LOT_ of time.
Its a lot more practical to decide on a target uptime which will be Good
Enough (tm) for a 0.5.2 release.
>Since freenet can't perform this function, I don't see the benefit of
>hosting .torrent files on freenet vs. a regular web server
the practical benefit is enormous because sites that host many .torrents
need a lot of bandwith; freenet will take care of that.
Also, while RIAA may still go after i
Instead of competing, lets cooperate! ;-))
It is entirely possible to run a freesite with .torrent links. It is
slightly incovenient because you have to save the .torrent file and
manually start BitTorrent. Can we make a special rule for .torrent
files in fproxy that displays a big fat warnin
after the fix of the latest nio bug I'm finally seeing succesful
announcements with nio node.
However, I noticed that announcement HTL is hardcoded to 15. The value
from freenet.conf/ini is never even parsed. Is there any particular
reason for that?
__
Ok, now I'm officially establishing a fund for hiring an assasin for the
sun nio engineers.
Their selector.select() call is screwed. you have to iterate through
the list of all registered keys and check each one. In plain English,
that means we were missing reads; I noticed in the logs sever
/
//
/>>/ The only issue that should be ironed out is that the rt infolet records
/>>/ both a failure and a success for a succesfull connection as this affects
/>>/ the CP rating somewhat:
/>>/ Consecutive Failures = 1
/>>/ Connection Attempts = 1
/>>/ Succesful Connections = 1
/>
I'll be fixing th
There seems to be a problem with authentication when nio nodes interact
with non-nio nodes. I and tessier tried a small network of 2 nio nodes,
and it worked just fine. (The LAN experiment obviously failed, probably
due to hostnames not resolving to the address in the ref or something)
The on
Hello,
Its just really unfortunate that we were told about NBIO so late in the
migration to non-blocking architecture; the NBIO api has some
not-so-trivial to overcome differences from java.nio; more specifically
the lack of ByteBuffers and attachments will require very major
infrastructural c
Since with nio we can afford to keep 256-512 connections open (and well
tuned uber-nodes can go way over 1000), we should rethink whether LRU is
still the best strategy for replacing existing connections. Currently
the connection that has had least activity is being dropped when a new
one is n
hello,
there is a new nio snapshot. quoting Matthew: "looks like it's probably
more or less working now"
http://freenetproject.org/snapshots/freenet-nio-061003.jar
if you want to test it, download it and save it as freenet.jar. There
are some relevant settings you may want to play with in
hi,
I've noticed this but haven't been able to investigate it thoroughly. Can
you update to recent CVS checkout (some changes happened to the local
Interface) and try running it again? The best test would be running it
for a while with FNP traffic disabled (on a computer not connected to the
int
Could we add an "impossible" level to the logger? It will be of lesser
priority than the Error level and will be used to log things that are
impossible to happen or at least we think they are impossible
___
devl mailing list
[EMAIL PROTECTED]
http://h
The migration to nio sockets is in cvs branch "nio". It is a simple
replacement with lots of workarounds and it is actually slower than the
original. However it was a necessary first step.
I would like one or two (but no more please) well-connected nodes to
give it a try. If there is output
well, I discovered two huge bugs in the java.nio package. The first one
I mailed to Matthew and it is that the read() reads bytes as signed
octets even though in the javadocs it explicitly says the values
InputStream.read() returns ranges 0-255. So for example 0xfb is -5, not
251. This will
well, I discovered two huge bugs in the java.nio package. The first one
I mailed to Matthew and it is that the read() reads bytes as signed
octets even though in the javadocs it explicitly says the values
InputStream.read() returns ranges 0-255. So for example 0xfb is -5, not
251. This will
Here is a tentative rollout of non-blocking polling for live
connections, assuming everything else (handshaking, authentication &
trailing field stuff) stays blocking. This should save 30-40% of
threads used. If authentication is also nio'd that percentage will rise
to above 50.
1. change al
Here is a tentative rollout of non-blocking polling for live
connections, assuming everything else (handshaking, authentication &
trailing field stuff) stays blocking. This should save 30-40% of
threads used. If authentication is also nio'd that percentage will rise
to above 50.
1. change al
Iff the native win32 fec libs are stable, why not include them with the
default installation? I'll test them on a windoze box for a few days
and if they are will host them on the frost website.
An item for the wishlist: native CHK calculation for win32. Right now
the biggest hog in the avera
Iff the native win32 fec libs are stable, why not include them with the
default installation? I'll test them on a windoze box for a few days
and if they are will host them on the frost website.
An item for the wishlist: native CHK calculation for win32. Right now
the biggest hog in the avera
I haven't slept all night but I got some preliminary thinking about how
nio will affect everything. In a nutshell, there's no way to confine
the changes to a single layer. Here's how I see it rolling out:
1. use a single selector for all network reads and have a single thread
do a blocking se
I haven't slept all night but I got some preliminary thinking about how
nio will affect everything. In a nutshell, there's no way to confine
the changes to a single layer. Here's how I see it rolling out:
1. use a single selector for all network reads and have a single thread
do a blocking se
Well,
I can't do a representative benchmark because my node is never on for
long and rarely has load beyond 2000, but so far the client VM (much as
I hate to admit it) performs better with incremental gc than server with
incremental. So far:
server + gc < client +gc
but
server - gc > client -
Well,
I can't do a representative benchmark because my node is never on for
long and rarely has load beyond 2000, but so far the client VM (much as
I hate to admit it) performs better with incremental gc than server with
incremental. So far:
server + gc < client +gc
but
server - gc > client -
Load balancing currently is designed to avoid overloading; current
statistics show great discrepancies between the requests different
nodes receive. Ian had calculated the stddev to be around 7000. I am
very interested in analyzing in depth possible additions and improvments
to the load bala
Load balancing currently is designed to avoid overloading; current
statistics show great discrepancies between the requests different
nodes receive. Ian had calculated the stddev to be around 7000. I am
very interested in analyzing in depth possible additions and improvments
to the load bala
AFAIK the current mechanism of providing seednodes to new nodes is to
make a dump of hawk's routing table at a given interval and provide that
as seednodes.ref. That way all new nodes get more or less the same set
of references. No single node's table is dynamic enough to provide
sufficient d
AFAIK the current mechanism of providing seednodes to new nodes is to
make a dump of hawk's routing table at a given interval and provide that
as seednodes.ref. That way all new nodes get more or less the same set
of references. No single node's table is dynamic enough to provide
sufficient d
Hi,
Since the new datastore can span across several files, wouldn't it be
slightly better for searching and fragmentation of the datastore to
distribute the stored keys across the volumes?
A very rough example would be to span the datastore on 16 files, and
store keys starting with 0 on the
There is lots of technical information in the old FAQ. It would be a
good idea to cut out what is obsolete and add that to a technical
section of the new FAQ, or maybe to the Architecure page.
I'm not confident enough in my knowledge of the internals to do it all
by myself, so I copied the ent
Hello,
in order to make it easier for people working on projects related to
Freenet to make their projects known I created a page in the public area
of the website.
Unlike the Third Party Tools page this one is oriented for those
projects that are new, need developers (delta407 !) or are stil
>Just about everyone in Europe, North America, and
much >of the Asia Pacific
>region can get such a connection if they want it bad
>enough.
wrong. Just about everyone in the above regions
except North America _cannot_ afford unNATted
broadband.
>Transferring data through a proxy is wasteful an
I remember long time ago (1993) there was a
compression program that had the feature of adding a
special checksum which was 2-5% of the size of the
file and then could restore a rather significan number
of missing or corrupt segments.
I remember personally zeroing out 100 disk sectors of
a file an
I remember long time ago (1993) there was a
compression program that had the feature of adding a
special checksum which was 2-5% of the size of the
file and then could restore a rather significan number
of missing or corrupt segments.
I remember personally zeroing out 100 disk sectors of
a file a
Sorry for the repost - I'm not subscribed and since
the terrorist attacks I haven't been able to connect
to lists.freentproject.org at all - the browser
resolves it to 66.27.255.141 and then dies so I can't
even check the archives..
can you guys please subscribe me to devl, tech and
support with
Sorry for the repost - I'm not subscribed and since
the terrorist attacks I haven't been able to connect
to lists.freentproject.org at all - the browser
resolves it to 66.27.255.141 and then dies so I can't
even check the archives..
can you guys please subscribe me to devl, tech and
support with
> I don't see what the issue is here. Separate
> attempts were made to
> negotiate a connection with the target node on
> different threads.
> Each failed, so each decrements the CP.
>
> --gj
Well I tried again with debug level and it still seems
that there is only one request made to the node,
> I don't see what the issue is here. Separate
> attempts were made to
> negotiate a connection with the target node on
> different threads.
> Each failed, so each decrements the CP.
>
> --gj
Well I tried again with debug level and it still seems
that there is only one request made to the node,
Hi Gianni,
I was looking at the minor log of build 405 and it
struck me that at random times the CP is reduced twice
due to an authentication error. Then I checked it
with the latest 404 and it behaved the same way. Here
are two portions of the log file. I'm sending this to
you since you added t
Hi Gianni,
I was looking at the minor log of build 405 and it
struck me that at random times the CP is reduced twice
due to an authentication error. Then I checked it
with the latest 404 and it behaved the same way. Here
are two portions of the log file. I'm sending this to
you since you added
101 - 195 of 195 matches
Mail list logo