Opened 9 months ago

Last modified 5 months ago

#33350 new defect

Is sbws weighting some relays too high?

Reported by: teor Owned by:
Priority: Medium Milestone: sbws: 1.1.x-final
Component: Core Tor/sbws Version:
Severity: Normal Keywords: sbws-roadmap, network-health
Cc: juga, gk, starlight Actual Points:
Parent ID: #33947 Points: 2
Reviewer: Sponsor:

Description

Before we deploy sbws to the rest of the bandwidth authorities, we should check if it is weighting some relays (or some ASes) much higher than torflow.

We should also check for bugs in sbws, that weight existing large ASes too high:
https://metrics.torproject.org/rs.html#aggregate/as

Child Tickets

Change History (8)

comment:1 Changed 9 months ago by teor

Keywords: network-health added

comment:2 Changed 9 months ago by gk

Cc: gk added

comment:3 Changed 8 months ago by gk

Parent ID: #33121

Setting as child to the "let's deploy sbws everywhere"-ticket according to the ticket description.

comment:4 Changed 7 months ago by juga

i think the main reason why this is happening is because:

  • #33831: relays without observed or average bandwidth in the descriptors are not scaled, but still added to the bwlines without vote=0
  • relays without that bandwidth in the descriptor but bandwidth burst (dunno why they have bandwidth burst and not the other, but i've seen it happening), would scale against bandwidth burst, which is higher

comment:5 Changed 7 months ago by juga

Since #33590 is a duplicate of this, adding here the info starlight added and closing it:

https://metrics.torproject.org/totalcw.png?start=2019-12-18

     tingpo                              
bastet---  3600 0.0150 +0.4605 0 39    
Faravahar  4980 0.0208 +1.0203 0 39    
maatuska- 2275652 9.4867 +922.1891 0 39

     NLRelay05                              
bastet---     1 0.0000 -0.9998 34 1747    
moria1---  5770 0.0239 +0.1819 34 1747    
maatuska- 1870361 7.7590 +382.1240 34 1747
Last edited 7 months ago by juga (previous) (diff)

comment:6 Changed 7 months ago by juga

Cc: starlight added

comment:7 Changed 6 months ago by juga

Parent ID: #33121#33947

Let's also check this as part of #33947

comment:8 Changed 5 months ago by juga

Last Saturday, 23th May, maatuska has updated to sbws version 1.1.0+84.g3033421 (current maint-1.1 branch) and can be observed that the total consensus bandwidth is decreasing close to the rest of bwauths.
i'm not closing this ticket yet cause i'd like to give an explanation on why i think this happening and because there are still some relays being measured too high (see https://trac.torproject.org/projects/tor/ticket/33947#comment:2)

Note: See TracTickets for help on using tickets.