LightningFilter: Traffic Filtering at 100 Gbps Presented by: - - PowerPoint PPT Presentation

lightningfilter traffic filtering at 100 gbps
SMART_READER_LITE
LIVE PREVIEW

LightningFilter: Traffic Filtering at 100 Gbps Presented by: - - PowerPoint PPT Presentation

LightningFilter: Traffic Filtering at 100 Gbps Presented by: Benjamin Rothenberger In collaboration with: Prof. Adrian Perrig, Juan Garca Pardo, Dominik Roos, Jonas Gude, Pascal Sprenger, Florian Jacky Project Goals High-speed packet


slide-1
SLIDE 1

LightningFilter: Traffic Filtering at 100 Gbps

Presented by: Benjamin Rothenberger

In collaboration with:

  • Prof. Adrian Perrig, Juan Garcìa Pardo, Dominik Roos,

Jonas Gude, Pascal Sprenger, Florian Jacky

slide-2
SLIDE 2

Project Goals

  • High-speed packet processing requires nanosecond
  • perations
  • Example: 64-byte packets @ 100Gbps: ~5ns processing time
  • Nanosecond scale key establishment
  • Nanosecond scale packet authentication
  • Trivia: how “long” is a nanosecond?
  • Answer: light travels about 30cm in 1ns
slide-3
SLIDE 3

AS C

Use Case: Network Error Message Authentication

Internet AS A

BR

AS B

BR

S

Client Server

C

BR BR

X !

Solution: DRKey! Problem: only short time frame à Only possible using symmetric cryptography !

slide-4
SLIDE 4

DRKey

  • Novel protocol based on symmetric cryptography
  • Intel AES-NI instructions enable key derivation within ~50 cycles

à Nanosecond scale!

  • Key computation is up to 3 times faster than DRAM lookup!

à Computing the key is faster than storing it in memory! à Foundation for many DDOS defense mechanisms

slide-5
SLIDE 5

DRKey Performance

Factor: ~ 1450x

slide-6
SLIDE 6

Lightning Filter

Traffic Filtering at 100 Gbps

slide-7
SLIDE 7

Overview

Standard Firewall Lightning Filter

Authenticated traffic SCION traffic normal traffic Invalid traffic Firewall traffic

Internet

Border Router

$$$

slide-8
SLIDE 8

System Design

L

............ ............ CONFIG FILE ............ ............

Metrics CLI DRKey Mgmt

Lightning Filter Data Plane Control Plane

Traffic Class. Source Auth. Rate Limiting Duplicate Supp.

Administrator Certificate Server Prometheus System Metric Exporter

slide-9
SLIDE 9

Demo Outline

  • 1. Attack scenario
  • Attacker located anywhere in Internet à Source authentication
  • 2. Bandwidth capacity
  • 120 Gbps traffic volumne
  • 3. Filtering based on source authentication
  • Alternate between filtering and bypass every 30s
  • 4. Duplicate suppression
  • 80 Gbps duplicates traffic, 40 Gbps legitimate traffic
slide-10
SLIDE 10

Attack Scenario: Internet Attacker

Internet

AS A

100 Mbps 120 Gbps 100 Mbps

slide-11
SLIDE 11

Attack Scenario: Internet Attacker

Internet

AS A

100 Mbps

LF

X

120 Gbps 100 Mbps

slide-12
SLIDE 12

Questions?

slide-13
SLIDE 13

Backup Slides

slide-14
SLIDE 14
  • Communication between clients and server is authenticated

using DRKey

  • Key derivation for L2 keys is delegated to server

DRKey Scenario

AS Server AS AS Clients

Internet

slide-15
SLIDE 15

DRKey Exchange Demo

  • 1. Client requests the L2 key to communicate to the server from its

local CS

  • 2. L1 key has not been prefetched à L1 key exchange
  • 3. Server fetches the derivation secret for its delegation from CS
  • 4. Server then derives the same L2 key locally
  • 5. Do 100 runs and calculate average execution time
slide-16
SLIDE 16

DRKey Hierarchy

  • Key establishment using a multi-level key hierarchy
  • L0: per-AS local secret key & per-AS public/private key pair
  • L1: AS-level key establishment (typically prefetched!)
  • L2: locally derive symmetric keys for end hosts
slide-17
SLIDE 17

DRKey Key Exchange

AS B Internet AS A

BR BR CS CS

C S

Clients Server

L1 key exchange Fetch DS C C1 Fetch L2 key Locally derive L2 Key

slide-18
SLIDE 18

Key Rollover

Key 𝐸𝑇#$% Fetching Key 𝐸𝑇#$& Active Key 𝐸𝑇# Key Rollover Grace Period Grace Period Fetching Key 𝐸𝑇#$' 0x0: 0x1: 0x2: Fetching Key 𝐸𝑇#$( t t + 1 t + 2 Key 𝐸𝑇#$% Key 𝐸𝑇#$' Active Key 𝐸𝑇#$' Active Key 𝐸𝑇#$%

slide-19
SLIDE 19

Rate Limiting

allocation for next silce Used tokens in last slice Refill rate I) aggregate c1 c0 c3 c2 c1 c0 c3 c2 c1 c0 c3 c2 II) recompute III) distribute

Data Plane

a) Packet processing