That was my original suggestion, with one small change.

Always try to upload the file with the earliest deadline, so that if a 
few uploads are occasionally going through, you get the most important 
one first.

-- Lynn

[email protected] wrote:
> A couple of suggestions.
> 
> 1)  When a backoff on the client ends, only start ONE upload.  (Not the
> maximum allowed by the client).
> 2)  If that upload completes, start the next ONE.
> 3)  If all upload with no problem, drop the backoff to 0.
> 4)  If one does NOT upload, continue with the exponential backoff.
> 
> jm7
> 
> 
>                                                                            
>              Martin                                                        
>              <[email protected]                                             
>              o.uk>                                                      To 
>              Sent by:                  BOINC Developers List               
>              boinc_dev-bounces         <[email protected]>        
>              @ssl.berkeley.edu                                          cc 
>                                                                            
>                                                                    Subject 
>              07/12/2009 08:06          Re: [boinc_dev] Optimizing          
>              PM                        uploads..... (AND downloads)        
>                                                                            
>                                                                            
>                                                                            
>                                                                            
>                                                                            
>                                                                            
> 
> 
> 
> 
> OK... To summarise/clarify from offlist...
> 
> (NB: Mb = Mega-bits. Mega-Bytes is MB)
> 
> 
> The assumption is that the core problem for the s...@h uploads/downloads
> being strangled is due to disgraceful link bandwidth degradation due to
> link congestion. This is highly suggestive when looking at the cricket
> graphs.
> 
> For example, at the moment the s...@h link shows:
> 
> http://fragment1.berkeley.edu/newcricket/grapher.cgi?target=%2Frouter-interfaces%2Finr-250%2Fgigabitethernet2_3;ranges=m;view=Octets
> 
> 
> Average bits in (for the day):
> Cur: 65.82 Mbits/sec
> Avg: 64.92 Mbits/sec
> Max: 74.95 Mbits/sec
> 
> Average bits out (for the day):
> Cur: 18.14 Mbits/sec
> Avg: 18.11 Mbits/sec
> Max: 21.31 Mbits/sec
> Last updated at Sun Jul 12 16:28:39 2009
> 
> And from the graph, earlier download peaks are pegged at about 93Mbit/s.
> Uploads peak at 20Mb/s.
> 
> Note that this is for a period where s...@h appear to be short of available
> WUs! (Or is this a supply throttle to test scenarios?)
> 
> 
> (Note also that the "bits in" "bits out" direction is confusingly
> reversed wrt the s...@h servers due to that router's reporting config.)
> 
> The present link load is 70Mb/s and connections from my Boinc client are
> fine and fast. Eg just now:
> 
> Mon 13 Jul 2009 00:38:55 BST         s...@home         Sending scheduler
> request:
> Requested by user.
> Mon 13 Jul 2009 00:38:55 BST         s...@home         Requesting new tasks
> Mon 13 Jul 2009 00:39:00 BST         s...@home         Scheduler request
> completed: got
> 8 new tasks
> Mon 13 Jul 2009 00:39:02 BST         s...@home         Started download of
> 28au08ag.27740.18477.4.8.252
> Mon 13 Jul 2009 00:39:02 BST         s...@home         Started download of
> 28au08ag.27740.18477.4.8.242
> Mon 13 Jul 2009 00:39:07 BST         s...@home         Finished download of
> 
> 28au08ag.27740.18477.4.8.252
> Mon 13 Jul 2009 00:39:07 BST         s...@home         Finished download of
> 
> 28au08ag.27740.18477.4.8.242
> Mon 13 Jul 2009 00:39:07 BST         s...@home         Started download of
> 15se08aa.19921.11115.9.8.138
> Mon 13 Jul 2009 00:39:07 BST         s...@home         Started download of
> 04dc08ae.32262.20522.4.8.62
> Mon 13 Jul 2009 00:39:12 BST         s...@home         Finished download of
> 
> 15se08aa.19921.11115.9.8.138
> Mon 13 Jul 2009 00:39:12 BST         s...@home         Finished download of
> 
> 04dc08ae.32262.20522.4.8.62
> Mon 13 Jul 2009 00:39:12 BST         s...@home         Started download of
> 28au08ag.27740.18477.4.8.254
> Mon 13 Jul 2009 00:39:12 BST         s...@home         Started download of
> 15se08aa.19921.11115.9.8.140
> Mon 13 Jul 2009 00:39:18 BST         s...@home         Finished download of
> 
> 28au08ag.27740.18477.4.8.254
> Mon 13 Jul 2009 00:39:18 BST         s...@home         Finished download of
> 
> 15se08aa.19921.11115.9.8.140
> Mon 13 Jul 2009 00:39:18 BST         s...@home         Started download of
> 04dc08ae.32262.20522.4.8.95
> Mon 13 Jul 2009 00:39:18 BST         s...@home         Started download of
> 04dc08ae.32262.20522.4.8.75
> Mon 13 Jul 2009 00:39:23 BST         s...@home         Finished download of
> 
> 04dc08ae.32262.20522.4.8.95
> Mon 13 Jul 2009 00:39:23 BST         s...@home         Finished download of
> 
> 04dc08ae.32262.20522.4.8.75
> 
> Nice 'n' quick. That is actually pegged at the maximum that my own
> traffic management permits (linux "tc", but NOT the "policer").
> 
> Sooo... A s...@h link average of 70Mb/s looks good :-)
> 
> 
> 
> To add the full list of proposed fixes (from myself and others):
> 
> Martin wrote:
> [...]
>> Server-side code can later be added to dynamically adjust the parameters
>> from learning about what happens on average.
> 
> For example, to control the link data rate from server side, dynamically:
> 
> Adjust the maximum number of simultaneous connections accepted;
> Adjust transfer rate limits for sending out data;
> Delay server responses and send NACK instead if still too many
> simultaneous connections.
> 
> 
>> That's the nature of a DDOS attack. That shouldn't be happening with the
>> present Boinc clients, and especially not with the exponential backoff.
> 
> Modify the existing client-side exponential backoff so that once a
> backoff time has accumulated, only reduce that backoff with a linear
> decay or better, a slow start exponential decay.
> 
> Allow the servers to update the client-side exponential backoff
> parameters depending on the project's current conditions. Send this in
> the initial response?
> 
> Always use ordered uploads: EDF if the last transfer completed
> successfully, RR if the last transfer failed (so that one rogue blocked
> result doesn't 'jam up the works'.
> 
> 
>> Very many more than the number of uploads/downloads that can be
>> accommodated down 90Mb/s of pipe to a modern world community of
>> broadband users. *Just 20 'average' users* simultaneously downloading
>> can easily saturate the downlink to a congested mess beyond what tcp can
>> handle. Similarly so for 150 'average' users all trying to upload during
>> the same one second.
>>
>> Those numbers are vastly smaller than 184320 and also a very much
>> smaller than the limits for apache.
> [...]
>> Hence:
>>
>>  >> As an experiment, can s...@h limit the max number of simultaneous
>>  >> upload/download connections to see what happens to the data rates?
>>  >>
>>  >> I suggest a 'first try' *max simultaneous connections of 150* for
>>  >> uploads *and 20 for downloads* Adjust as necessary to keep the link
>>  >> at an average that is no more than *just 80 Mbit/s*
>>
>> It would be interesting to see what those numbers adjust to so as to hit
>> the unsaturated 80 Mbit/s... And see by how much that improves the
>> sustained transfer rate for the WUs.
> 
> Is that being tried at the moment?
> 
> And for what numbers?
> 
> Whatever has been done for the moment, at 70Mb/s the s...@h link is running
> sweet and smooth and fast!
> 
> 
> Regards,
> Martin
> 
> --
> --------------------
> Martin Lomas
> m_boincdev ml1 co uk.ddSPAM.dd
> --------------------
> _______________________________________________
> boinc_dev mailing list
> [email protected]
> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
> To unsubscribe, visit the above URL and
> (near bottom of page) enter your email address.
> 
> 
> 
> _______________________________________________
> boinc_dev mailing list
> [email protected]
> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
> To unsubscribe, visit the above URL and
> (near bottom of page) enter your email address.
> 
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to