qi huang, ken birman, robbert van renesse (cornell), wyatt lloyd (princeton, facebook), sanjeev...
TRANSCRIPT
![Page 1: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/1.jpg)
Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook),
Sanjeev Kumar, Harry C. Li (Facebook)
An Analysis ofFacebook Photo Caching
![Page 2: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/2.jpg)
2
250 Billion* Photos on Facebook
Profile
Feed
Album
* Internet.org, Sept., 2013
StorageBackend
CacheLayers
Full-stackStudy
![Page 3: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/3.jpg)
3
Preview of Results
Current Stack Performance
Opportunities for Improvement
• Browser cache is important (reduces 65+% request )
• Photo popularity distribution shifts across layers
• Smarter algorithms can do much better (S4LRU)
• Collaborative geo-distributed cache worth trying
![Page 4: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/4.jpg)
4
Client
Facebook Photo-Serving Stack
![Page 5: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/5.jpg)
5
Client-based Browser CacheClient
Browser
Cache
Client
LocalFetch
![Page 6: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/6.jpg)
6
Client-based Browser CacheClient
Browser
Cache(Millions)
Client
![Page 7: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/7.jpg)
7
Stack Choice
Browser
Cache
Client
CachesStorage
Facebook Stack
Akamai
Content Distribution Network(CDN)
• Focus: Facebook stack
(Millions)
![Page 8: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/8.jpg)
8
Geo-distributed Edge Cache (FIFO)
Edge Cache
(Tens)
Browser
Cache
Client
PoP
(Millions)
![Page 9: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/9.jpg)
9
Geo-distributed Edge Cache (FIFO)
Edge Cache
(Tens)
Browser
Cache
Client
PoP
(Millions)
Purpose
1. Reduce cross-country latency
2. Reduce Data Center bandwidth
![Page 10: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/10.jpg)
10
Geo-distributed Edge Cache (FIFO)
Edge Cache
(Tens)
Browser
Cache
Client
PoP
(Millions)
![Page 11: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/11.jpg)
11
Geo-distributed Edge Cache (FIFO)
Edge Cache
(Tens)
Browser
Cache
Client
PoP
(Millions)
![Page 12: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/12.jpg)
12
Single Global Origin Cache (FIFO)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
(Tens)(Millions) (Four)
![Page 13: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/13.jpg)
13
Single Global Origin Cache (FIFO)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
(Tens)(Millions) (Four)
Purpose
1. Minimize I/O-bound operations
![Page 14: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/14.jpg)
14
Single Global Origin Cache (FIFO)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
(Tens)(Millions) (Four)
Hash(url)
![Page 15: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/15.jpg)
15
Single Global Origin Cache (FIFO)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
(Tens)(Millions) (Four)
![Page 16: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/16.jpg)
16
Haystack Backend
Backend (Haystack
)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
(Tens)(Millions) (Four)
![Page 17: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/17.jpg)
17
How did we collect the trace?
![Page 18: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/18.jpg)
18
Trace Collection
Instrumentation Scope
Backend (Haystack
)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
(Object-based sampling)
• Request-based: collect X% of requests• Object-based: collect reqs for X% objects
![Page 19: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/19.jpg)
19
Sampling on Power-law
Object rank
![Page 20: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/20.jpg)
20
Sampling on Power-law
• Req-based: bias on popular content, inflate cache perf
Object rank
Req-based
![Page 21: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/21.jpg)
21
Sampling on Power-law
Object-based
Object rank
• Object-based: fair coverage of unpopular content
![Page 22: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/22.jpg)
22
Sampling on Power-law
• Object-based: fair coverage of unpopular content
Object rank
Object-based
![Page 23: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/23.jpg)
23
Trace Collection
77.2M reqs
(Desktop)
12.3MBrowser
s
Instrumentation Scope1.4M photos, all reqs for each
Backend (Haystack
)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
Resizer
R
2.6M photo objects, all reqs for each
12.3KServers
![Page 24: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/24.jpg)
24
Analysis
• Traffic sheltering effects of caches
• Photo popularity distribution
• Size, algorithm, collaborative Edge
• In paper– Stack performance as a function of photo age– Stack performance as a function of social
connectivity– Geographical traffic flow
![Page 25: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/25.jpg)
25
Traffic Sheltering
77.2M
26.6M11.2M
7.6M
Backend (Haystack
)
Browser
Cache
Edge Cache
OriginCache
PoPClient
Data Center
65.5%58.0%
31.8%
R
Traffic Share
65.5% 20.0% 4.6% 9.9%
![Page 26: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/26.jpg)
26
Photo popularity and its cache impact
![Page 27: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/27.jpg)
27
Popularity Distribution
• Browser resembles a power-law distribution
2%
![Page 28: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/28.jpg)
28
Popularity Distribution
• “Viral” photos becomes the head for Edge
![Page 29: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/29.jpg)
29
Popularity Distribution
• Skewness is reduced after layers of cache
![Page 30: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/30.jpg)
30
Popularity Distribution
• Backend resembles a stretched exponential dist.
![Page 31: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/31.jpg)
31
Popularity with Absolute Traffic
• Storage/cache designers: pick a layer
![Page 32: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/32.jpg)
32
Popularity Impact on Caches
High Low M
Lowest
Each has 25% requests
![Page 33: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/33.jpg)
33
Popularity Impact on Caches
• Browser traffic share decreases gradually
![Page 34: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/34.jpg)
34
Popularity Impact on Caches
• Edge serves consistent share except for the tail
7.8%
22~23%
![Page 35: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/35.jpg)
35
Popularity Impact on Caches
• Origin contributes most for “low” group
9.3%
![Page 36: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/36.jpg)
36
Popularity Impact on Caches
• Backend serves the tail
70% Haystack
![Page 37: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/37.jpg)
37
Can we make the cache better?
![Page 38: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/38.jpg)
38
Simulation
• Replay the trace (25% warm up)
• Estimate the base cache size
• Evaluate two hit-ratios (object-wise, byte-wise)
![Page 39: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/39.jpg)
39
Edge Cache with Different Sizes
• Picked San Jose edge (high traffic, median hit ratio)
59%
![Page 40: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/40.jpg)
40
Edge Cache with Different Sizes
• “x” estimates current deployment size (59% hit ratio)
65%68%
59%
![Page 41: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/41.jpg)
41
Edge Cache with Different Sizes
• “Infinite” size ratio needs 45x of current capacity
Infinite Cache
65%68%
59%
![Page 42: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/42.jpg)
42
Edge Cache with Different Algos
• Both LRU and LFU outperforms FIFO slightly
Infinite Cache
![Page 43: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/43.jpg)
43
S4LRU
Cache Space
More Recent
L3
L2
L1
L0
![Page 44: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/44.jpg)
44
S4LRU
Cache Space
L3
L2
L1
L0Missed Object
More Recent
![Page 45: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/45.jpg)
45
S4LRU
Cache Space
L3
L2
L1
L0
HitMore Recent
![Page 46: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/46.jpg)
46
S4LRU
Cache Space
L3
L2
L1
L0
Evict
More Recent
![Page 47: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/47.jpg)
47
Edge Cache with Different Algos
• S4LRU improves the most
68%
1/3x
Infinite Cache
59%
![Page 48: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/48.jpg)
48
Edge Cache with Different Algos
• Clairvoyant (Bélády) shows much improvement space
Infinite Cache
![Page 49: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/49.jpg)
49
Origin Cache
• S4LRU improves Origin more than Edge
14%
Infinite Cache
![Page 50: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/50.jpg)
50
Which Photo to Cache
• Recency & frequency leads S4LRU
• Does age, social factors also play a role?
![Page 51: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/51.jpg)
51
Collaborative cache on the Edge
![Page 52: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/52.jpg)
52
Geographic Coverage of Edge
Small working set
![Page 53: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/53.jpg)
53
Geographic Coverage of Edge
9 Edges with high-volume traffic
![Page 54: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/54.jpg)
54
Geographic Coverage of Edge
Do clients stay with local Edge?
![Page 55: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/55.jpg)
55
Geographic Coverage of Edge
Atlanta
![Page 56: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/56.jpg)
56
Geographic Coverage of Edge
Atlanta
20% local
5% Dallas
35% D.C.
5% NYC
20% Miami
5% California
10% Chicago
• Atlanta has 80% requests served by remote Edges
![Page 57: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/57.jpg)
57
Geographic Coverage of Edge
Atlanta
20% local
Miami
35% localDall
as50% local
Chicago
60% local
LA 18% local
NYC 35% local
• Substantial remote traffic is normal
![Page 58: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/58.jpg)
58
Geographic Coverage of Edge
Amplified working set
![Page 59: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/59.jpg)
59
Collaborative Edge
![Page 60: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/60.jpg)
60
Collaborative Edge
• “Independent” aggregates all high-volume Edges
![Page 61: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/61.jpg)
61
Collaborative Edge
• “Collaborative” Edge increases hit ratio by 18%
18%
Collaborative
![Page 62: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/62.jpg)
62
Related WorkStorage Analysis
Content Distribution Analysis
Web Analysis
BSD file system (SOSP ’85), Sprite ( SOSP ’91), NT (SOSP ’99),NetApp (SOSP ’11), iBench (SOSP ’11)
Cooperative caching (SOSP ’99), CDN vs. P2P (OSDI ’02),P2P (SOSP ’03), CoralCDN (NSDI ’10), Flash crowds (IMC ’11)
Zipfian (INFOCOM ’00), Flash crowds (WWW ’02),Modern web traffic (IMC ’11)
![Page 63: Qi Huang, Ken Birman, Robbert van Renesse (Cornell), Wyatt Lloyd (Princeton, Facebook), Sanjeev Kumar, Harry C. Li (Facebook) An Analysis of Facebook Photo](https://reader036.vdocument.in/reader036/viewer/2022062511/551677715503469d698b5a31/html5/thumbnails/63.jpg)
63
Conclusion
• Quantify caching performance
• Quantify popularity changes across layers of caches
• Recency, frequency, age, social factors impact cache
• Outline potential gain of collaborative caching