On Tue, 2003-11-25 at 07:36, Ed Tomlinson wrote:
> My node is accepting about 10% of requests.  In my mind this it is not
> working well at all.  I think the backoff scheme in use is using alchemy.
> ie. there is no predictable result - we just say it is working...  
> 

My node has been accepting around 59% of requests for the past 12 days.
Just before that, it was accepting only a few percent.

(I am still blacklisting nodes that do not backoff, by IP at the
firewall, but I don't think that matters much now).

I have 1.5gb RAM, 2GHz P4, DSL.  Configuration parameters:
threadFactory=Y
storeSize=4096M
outputBandwidthLimit=5400
maxNodeConnections=800
maximumThreads=500
defaultResetProbability=0.01

Every so often (maybe every 10 minutes) it drops to accepting only
a few percent of requests, because of messageSendTimeRequest, but
quickly recovers.

Over the past 2 days, I averaged 9 nodes backed off (24 samples),
max observed 28, out of 50 nodes in the RT.  Recently I haven't
seen any long backoff times.

Since the inception of backoff, I averaged 17 nodes backed off 
(123 samples), max 34.

Total pooled threads averaged 191 over past 2 days, and 128 since Sept.
Largest number caught (which is probably not the maximum) was 370.

Total data received is almost always more than data sent, but I
note that those numbers are really data + overhead, not just trailers.

outputBytes is often about 5 times outputBytesTrailingAttempted.

Bytes waiting to be sent recent average is 100 MiB. 
That takes about 5 hours to send at 6k/sec.  

Recent probability of success of an incoming request Max (out of 16
bins) averaged 2.5%.  Since Sept. it averaged 3%.

Does this mean routing is not working?  Maybe.  Maybe it means 97%
of requests are for nonexistent keys.  Or maybe something in between.

-- Ed Huff

Attachment: signature.asc
Description: This is a digitally signed message part

_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to