bugThe Circle - Bugs: bug #3648, slow peer can slow things down

 
 

bug #3648: slow peer can slow things down

Submitted by:  Paul Harrison <pfh>
Submitted on:  Sat 17 May 2003 02:08:24 AM UTC  
 
Category: InternalSeverity: 3 - Ordinary
Item Group: BugStatus: Fixed
Privacy: PublicAssigned to: None
Open/Closed: Closed

Add a New Comment(Rich Markup)
   

You are not logged in

Please log in, so followups can be emailed to you.

 

Fri 02 Jan 2004 02:45:19 PM UTC, comment #4:

fixed in the generators version

ThomasV <thomasV>
Project Administrator
Tue 01 Jul 2003 08:46:53 AM UTC, comment #3:

I modified the hashtable activation decision, so that a peer who becomes slow after it has activated its hashtable deactivates it within a reasonable time.
(afaik a peer can become temporarily slow, depending on what it does with its connection, eg if it runs ftp or other p2p apps)

ThomasV <thomasV>
Project Administrator
Sat 28 Jun 2003 04:24:14 PM UTC, comment #2:

peers are probed in node.find_nodes.
I ran some statistics to know often this is done for each peer.
It turns out that the probes are very unevenly distributed among peers. This is due to the way the list variable is updated. Some peers get probed hundreds of times, while others only 1 or 2. pings_over_threshold and ping_under_threshold are very sensitive to this.
It would make more sense to estimate the pingtime of each peer, and to activate only if :
peers_having_pingtime_under_threshold<peers_having_pingtime_under_threshold

ThomasV <thomasV>
Project Administrator
Sat 28 Jun 2003 10:26:01 AM UTC, comment #1:

peers are probed in node.find_nodes.
I ran some statistics to know often this is done for each peer.
It turns out that the probes are very unevenly distributed among peers. This is due to the way the list variable is updated. Some peers get probed hundreds of times, while others only 1 or 2. pings_over_threshold and ping_under_threshold are very sensitive to this.
It would make more sense to estimate the pingtime of each peer, and to activate only if :
peers_having_pingtime_under_threshold<peers_having_pingtime_under_threshold

ThomasV <thomasV>
Project Administrator
Sat 17 May 2003 02:08:24 AM UTC, original submission:

A slow peer on the network can apparently slow everyone else down. Normally a slow peer would not activate it's hashtable, possibly this code is buggy. However it would be nice to cope with slow peers in any case.

One problem is the overloading avoidance stuff. If Circle started as many threads as it wanted, queries with redundancy would return results quickly even if one of the relevant nodes was slow. However if there are limited threads, they can all be tied up talking to one slow node.

Another problem is that Node likes to do a batch of tasks, wait for all of them to finish, then put together the next batch, so a single slow task can slow Node down.

Paul Harrison <pfh>
Project Administrator

 

(Note: upload size limit is set to 16384 kB, after insertion of the required escape characters.)

Attach File(s):
   
   
Comment:
   

No files currently attached

 

Depends on the following items: None found

Items that depend on this one: None found

 

CC list is empty

 

Do you think this task is very important?
If so, you can click here to add your encouragement to it.
This task has 0 encouragements so far.

Only logged-in users can vote.

 

Please enter the title of George Orwell's famous dystopian book (it's a date):

 

 

Follow 2 latest changes.

Date Changed By Updated Field Previous Value => Replaced By
Fri 02 Jan 2004 02:45:19 PM UTCthomasVStatusNone=>Fixed
  Open/ClosedOpen=>Closed

Back to the top


Powered by Savane 3.1-cleanup1