Scaling the Netflix Global CDN - by Dave Temkin
An impressive and detailed presentation by Mr. Dave Temkin on how Netflix deploys caching nodes and challenges around them. It’s from Feb last year.
An impressive and detailed presentation by Mr. Dave Temkin on how Netflix deploys caching nodes and challenges around them. It’s from Feb last year.
One of very cool features of IPv6 is link-local address which stays local to a given link. For this fe80::/10 is reserved. A /10 is a huge amount of address space in IPv6 (and in IPv4 too :) ). This means from fe80:0000:0000:0000:0000:0000:0000:0000 to febf:ffff:ffff:ffff:ffff:ffff:ffff:ffff
.
Since by design link-local address stays local, the address configured on the upstream/gateway router can be kept same for ease of use and comfort. This wasn’t the case of IPv4 where each VLAN/layer 2 domain had it’s own gateway. So e.g if you have two VLANs or interfaces say: Gi1/0 and Gi2/0. You decide to use 10.100.100.0/30 on Gi1/0 and 10.100.100.4/30 on Gi2/0.
Today I spotted some routes from Amazon AWS Cloud services - AS16509 in Indian tables. AS16509 was originating prefixes while sitting in downstream of Tata-VSNL AS4755 and Reliance AS18101. I almost missed Amazon AWS's announcement on their blog about Indian PoPs for their DNS service - Route53 and CDN service - Cloudfront.
New PoP’s of Amazon in India are at Mumbai and Chennai and I see pretty much consistent BGP announcements to Tata and Reliance from these locations. Prefixes I have seen so far: