Midgress-aware traffic provisioning for content delivery Aditya - PowerPoint PPT Presentation

Midgress-aware traffic provisioning for content delivery Aditya Sundarrajan, Mangesh Kasbekar, Ramesh K. Sitaraman, Samta Shukla

CDNs serve more than 50% of content Midgress Egress Request / Response Origin Users CDN 2 2

Performance and cost metrics End-user latency Bandwidth cost Origin offload ratio Cache hit rate 3

100s of content providers 100s of 1000s of servers Millions of users 4

Cache Traffic management provisioning Traffic classes Requests Users 5

Cache Traffic management provisioning Traffic classes Requests Users Past work has focused on cache management 6

Cache Traffic management provisioning Traffic classes Requests Users How can we assign traffic classes to reduce midgress? 7

Traffic provisioning to reduce midgress S 1 + + Midgress x 1/2 S 2 x 1/2 100s of traffic assignment scenarios! 8

Traffic provisioning to minimize midgress Traffic classes Min. miss traffic to origin Optimize traffic class assignment Origin CDN Users 9

Eviction age equality Insert o 1 o 2 o 4 o 3 Evict …… Eviction age Head Tail 10 10

Footprint descriptors* Stack distance Spatial locality : How many unique Joint probability bytes are requested between distribution successive requests of an object? P(s,t) Inter-arrival time Temporal locality : How often is an object requested? 11 * Footprint descriptors: Theory and practice of cache provisioning in a global CDN, A. Sundarrajan et al. in ACM CoNEXT 2017

Caching properties from FDs Stack distance Cache size Hit rate Joint probability distribution P(s,t) Eviction age Cache size Hit rate = f (size) Cache size = f (eviction Inter-arrival time age) Hit rate Hit rate = f (eviction age) Eviction age 12 * Footprint descriptors: Theory and practice of cache provisioning in a global CDN, A. Sundarrajan et al. in ACM CoNEXT 2017

Traffic mixing using FD calculus FD 1 + FD 1+2 FD 2 The addition operation is the convolution of joint pdfs which can be efficiently computed using FFT 13 * Footprint descriptors: Theory and practice of cache provisioning in a global CDN, A. Sundarrajan et al. in ACM CoNEXT 2017

Traffic provisioning to minimize midgress FD of traffic Min. classes miss traffic to origin FD calculus to optimize traffic class Origin CDN Users assignment 14 14

Traffic provisioning as an optimization problem MILP – NP Hard!! T traffic classes λ 1, λ 2, …, λ T ∑ j x 1 j λ j … ∑ j x N j λ j Estimate Cache size, C miss rate of traffic 1 2 N -1 N Traffic mix using N servers capacity, B FD calculus Min. ∑ ij x ij λ j m j (c ij ) Total miss traffic from cluster 15

FD-based local search is faster than MILP 1. Randomly assign traffic classes Traffic classes … 1 2 N -1 N Servers Predict midgress of traffic mix using FD calculus 16

FD-based local search is faster than MILP 2. Reassign traffic classes using Traffic classes local search such that midgress is minimized … 1 2 N -1 N Servers Predict midgress of traffic mix using FD calculus 17

Metro-level traffic provisioning Traffic classes Cluster 1 Cluster N … … … Servers Servers Midgress of metro area 18

Trace characteristics Number of traffic classes 25 Length of trace 16 days Traffic types Web, media, download 19

Metro-level midgress reduced by 20% 60 50 Cache miss rate, % 40 OPT local search 30 baseline fit 20 10 0 0 100 200 300 400 500 600 Cache size, TB 20

Traffic provisioning in partitioned caches 60 50 OPT-share OPT Cache miss rate, % baseline fit-share baseline fit 40 OPT-part OPT – part baseline fit – part baseline fit-part 30 20 10 0 0 100 200 300 400 500 600 Cache size, TB 21

Conclusions Midgress-aware traffic provisioning reduced midgress by almost 20% in metro area Midgress-aware heuristic performs within 1.1% of OPT but is much faster Midgress-aware traffic provisioning can be extended to work with additional constraints such as minimum redundancy and maximum midgress, any cache management algorithm, and with partitioned caches 22

Th Thank k you! Em Email: asundar@cs.umass.edu 23