Exact Amount of Congestion S ERHAT A RSLAN , N ICK M C K EOWN Buffer - - PowerPoint PPT Presentation

β–Ά
exact amount of congestion
SMART_READER_LITE
LIVE PREVIEW

Exact Amount of Congestion S ERHAT A RSLAN , N ICK M C K EOWN Buffer - - PowerPoint PPT Presentation

Jan 19, 2024 46 likes β€’166 views

Switches Know the Exact Amount of Congestion S ERHAT A RSLAN , N ICK M C K EOWN Buffer Sizing Workshop, Stanford University Dec 2-3, 2019 Main Takeaways 1. We should directly measure the cause of congestion. 2. Congestion is buildup of the

slide-1
SLIDE 1

Switches Know the Exact Amount of Congestion

Buffer Sizing Workshop, Stanford University Dec 2-3, 2019

S ERHAT A RSLAN, N ICK M CK EOWN

slide-2
SLIDE 2

Serhat Arslan, sarslan@stanford.edu

Main Takeaways

  • 1. We should directly measure the cause of congestion.
  • 2. Congestion is buildup of the queue at a bottleneck link.
  • 3. Switches today can provide their buffer occupancy freely.
  • 4. Bottleneck buffer occupancy can be collected by updating a header

field on packets.

slide-3
SLIDE 3

Serhat Arslan, sarslan@stanford.edu

slide-4
SLIDE 4

Serhat Arslan, sarslan@stanford.edu

slide-5
SLIDE 5

Serhat Arslan, sarslan@stanford.edu

Why don’t switches simply tell everything?

Measure the congestion directly at the location of congestion itself…

slide-6
SLIDE 6

Serhat Arslan, sarslan@stanford.edu

Congestion Signal on TCP Header

slide-7
SLIDE 7

Serhat Arslan, sarslan@stanford.edu

π’ˆ( 𝒓 ) = πœ₯{ βˆƒπ’‹π’“π’‹ > π’–π’Šπ’”π’‡π’•π’Š }

Construct Your Custom Signal

π‘Ÿ = π‘Ÿ2 , … , π‘Ÿ56 , π‘₯β„Žπ‘“π‘ π‘“ 𝑙 β‰… β„Žπ‘π‘ž_π‘‘π‘π‘£π‘œπ‘’ π’ˆ( 𝒓 ) = βˆ‘π’‹E𝟐

πŸ‘π’ 𝒓𝒋

π’ˆ 𝒓 = π’π’ƒπ’š( 𝒓 ) π’ˆ( 𝒓 ) = βˆ‘π’‹E𝟐

πŸ‘π’ π’“π’‹πŸ‘

ECN RTT

Bottleneck!

slide-8
SLIDE 8

Serhat Arslan, sarslan@stanford.edu

Effectiveness Analysis

Ablation Scenario:

slide-9
SLIDE 9

Serhat Arslan, sarslan@stanford.edu

Effectiveness Analysis

slide-10
SLIDE 10

Serhat Arslan, sarslan@stanford.edu

Feasibility Analysis

Overhead Costs: Enough to collect 8 M

NOPQ RST β‰ˆ 0.4% of MTU

Latency Throughput

slide-11
SLIDE 11

Serhat Arslan, sarslan@stanford.edu

Main Takeaways

  • 1. We should directly measure the cause of congestion.
  • 2. Congestion is buildup of the queue at a bottleneck link.
  • 3. Switches today can provide their buffer occupancy freely.
  • 4. Bottleneck buffer occupancy can be collected by updating a header

field on packets.

slide-12
SLIDE 12

Questions?

S E R H A T A R S L A N s a r s l a n @ s t a n f o r d . e d u