1. CDN CACHE DISTRIBUTION
THROUGH RINEX
CASE STUDY: AKAMAI @ RINEX
AFPIF 2015, BY GHISLAIN
NKERAMUGABA
24-28 AUG, MAPUTO, MOZAMBIQUE
August 24-28, 2015 AfPIF 2015, MAPUTO - MOZAMBIQUE 1
2. BACKGROUND
• RINEX – Rwanda Internet Exchange AS37224
• Officially managed by RICTA since FEB 2014
• RICTA manages as well the RW ccTLD since
SEP 2012
• RICTA is a not-for profit private organization,
limited by guarantee
August 24-28, 2015 AfPIF 2015, MAPUTO - MOZAMBIQUE 2
3. BACKGROUND
• AKAMAI, approached during AFPIF 2013
• AKAMAI represented a HUGE portion of
“imported” content within Rwanda
• AVG speed: 350-500 Mbps
• The problem/challenge: Even if we bring
AKAMAI, HOW RINEX WILL FINANCE THE
TRANSIT LINK?
August 24-28, 2015 AfPIF 2015, MAPUTO - MOZAMBIQUE 3
4. THE SOLUTION
• RINEX to officially manage the AKAMAI cluster,
BUT to “host” it within an ISP [one RINEX peer]
• Tender: Winner = lowest $$/Mbps of cache [paid
peering model]
• Test phase: to evaluate transit ratios/levels
• AKAMAI cache content to ONLY be redistributed
through RINEX
• RINEX WOULD ONLY GET $$$ on port fees [IX
model]
August 24-28, 2015 AfPIF 2015, MAPUTO - MOZAMBIQUE 4
5. TECHNICAL ASPECTS
AfPIF 2015, MAPUTO - MOZAMBIQUE 5
4.86
350 Mbps 1.2 Gbps
AKAMAI
Cluster
Transit Link
INTERNET
KEY ASPECTS:
1. The Port speed capacity should accommodate traffic volume (AKAMAI +
Other)
2. ISP #0 is responsible in availing traffic statistics to RICTA (Supervision),
and to all ISPs that are accessing AKAMAI cache content
3. Distributing AKAMAI cache content is done ONLY through RINEX
4. ISP #0 shall only advertise to AKAMAI the LOCAL (to RINEX)
ASN/Prefixes
ISP
#1
ISP
#6
ISP
#5
ISP
#4
ISP
#3
ISP
#2
ISP
#X
ISP
#0
ISP #0
Customers
1
RINEX
Switch
August 24-28, 2015
6. RESULTS
• ISP #1: $80/Mbps
• ISP #2: $19/Mbps
• ISP #3 [BSC]: $12/Mbps: Content to be
distributed under a paid peering model
• RINEX Value/Traffic has increased
• RINEX keeps its operating model (*Switching*)
• End-user experience has improved
• Savings: ISPs are sharing the cost of “importing”
content (less cash outflows)
August 24-28, 2015 AfPIF 2015, MAPUTO - MOZAMBIQUE 6
7. OTHER EXPERIENCE: UG-GGC
• Two GGC instances in the country;
• At the beginning, GGC cache distributed *freely*
through UIXP;
• As consumption & Operators increased, Transit
link increased at the *hosting* operator (i.e. Cost)
• Hosting operator incurred high costs, eventually
stopped the service: The model proved to be
unsustainable
August 24-28, 2015 AfPIF 2015, MAPUTO - MOZAMBIQUE 7