14
Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Embed Size (px)

Citation preview

Page 1: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Internet Routing Instability

OFFENSE

Scott Crosby Atul Singh

Page 2: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Critique

• Poor title

• Level of contribution

• Clarity of presentation

• Correctness

• Methodology

• Claims

• Relevance

Page 3: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Poor Title

• Not about stability– But, its about routing traffic – Doesn’t tell desirable updates from undesirable

updates

• Better title : Characterization of BGP routing traffic

Page 4: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Desirableand Necessary

TopologyPolicy

Change nothing

Route updates

Change state

Undesirable

Page 5: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Contribution

• Measurement results must be clear for this to be a contribution

• Most paths are stable– Already known from Paxon[96]

• Free debugging– bad implementations– vendor screw ups

Page 6: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Clarity

• Figure 2– Only shows 10% of the picture – Where is

WWDup?– Appears uncorrected for missing data

• Useless to estimate true distribution

– Can’t eyeball ratio of different categories

Page 7: Internet Routing Instability OFFENSE Scott Crosby Atul Singh
Page 8: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Clarity

• Figure 3– Impressive graph with ~30,000 sample points– Unreadable, unnecessary – Only discusses trends – should graph trends.

Page 9: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Fig 3 : Internet forwarding instability density measure at Mae-eastExchange point during 1996.

Page 10: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Correctness

• '99% of routing messages are pathological'?– No data to backup their claim

• Analysis of routing messages– Where’s the table of the number of each type?– Where’s the table estimating the number of

unnecessary routing messages?

Page 11: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Methodology

• Only 5 sites, why is that enough? – Discount private exchanges

• Do the instabilities actually effect performance? – Claim yes, but no evidence– Ugliness is not a crime

• Uses prefix-pair tuple– How to deal with aggregation?

Page 12: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Claims

• ‘routing instability contributes to poor end-to-end performance?’– Paxon[96] paper shows 1% problems

• ‘high levels of instability can lead to packet loss’ – Paxon[96] shows this is rare– Unverified

Page 13: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

Relevance

• Does instability matter? – If the network changes, it changes

– A network with static routes would be perfectly stable• But not robust

• Debug vendor X’s code• Obsolescent

– new routers contain a full forwarding table in RAM

• Care about end-to-end performance

Page 14: Internet Routing Instability OFFENSE Scott Crosby Atul Singh

It’s a dirty job to critique but someone had to do it

• Poor title

• Level of contribution

• Clarity of presentation

• Correctness

• Methodology

• Claims

• Relevance