News Feed
The Content Providers Speak: At high level, what does Yahoo’s network look like? Data centers around the world (owner-operator to managed), delivers end-services portfolio. High focus on performance to end users. Vijay: volume driven, most services are free; network is bifurcating into express network: optical costs for data center to data center connectivity, and shorter runs. Dan: challenges are in finding ways to improve performance. We take up costs to improve the experience, no data centers outside the US. Some content doesn’t allow us to leverage 3rd party CDN, costs are also a consideration. Facebook is static content, majority of use is nano-content, not as expansible; we look at ways to accelerate, use of small pods with great amount of transport. Vijay: Power (electricity) is dominant cost, so data centers are located where energy is cheap, far from users. Optimizing for performance not same as optimizing for cost (backend). We want to bring services close to people, so content distribution network inside larger network. Raj: Yahoo is proud of making user experience relevant to user. Strategy: security management requires an ecosystem approach, all the things talked about with security overlay. What’s on my mind: maturing the infrastructure. Obviously there’s growth that drives attention from content providers. Dan: look at transport costs, look at Asia and new markets, we’ll pay what it takes (responsible costs). Users will come. Some markets, e.g., Hong Kong, are 2x more expensive, hard to justify costs so less pops.
Lease space from colocation operators? or lease commercial data space? Facebook does lease but at point now, 3-6MW of power required, where we need to build. Raj: largely the scale drives us to fewer data centers with lower energy footprints. This might not make commercial sense for colocation facilities. Border is around 3-4MW for economics. Vijay: we’re not building data centers, we’re building systems. Data center costs are divorced from machine pool. Yahoo DC2 model is very efficient. At these power levels, it’s a substantial footprint in commercial space.
Traffic exchange vs IP transit? Dan: at this point 50-50, not a cost view. It gets back to performance. Connect as much as possible with user ISPs. Raj: IP providers are significant part of our ecosystem, distribution over peering and IP provides diversity in cases of backbone outages and other challenges. Fragmented IP market challenges? There are mechanisms that allow aggregation, other approaches.
What services does Google buy: Carriers are partners. We work to make it cheaper for both to exchange traffic (video) for what users want and what we’re supplying (experience). Performance and upgradability: sometimes we need to go ahead of upgrade cycle and peering connections: where is the traffic and demand? Not a hard and fast rule to determine what’s best for carriers and users. Dan: we buy services once we meet a threshold, but doubling traffic every 6 months so need to turn up services frequently (every 30 days). We sit down and talk with providers, but concern that we’re eventually going to run out of time. Looking for alternative means to get around this.
Raj: focus of conference, cloud computing, front-end is moving closer to user, would love to build a non-blocking networks. Industry should think about representative architectures that can offer this non-blocking network at delivery points to address coming tsunami of needs. Large percentage of network assets are idle at given times, is there a better model that allows increase in efficiency in network? What kind of propositions will help this? Vijay: we’re still stuck with existing modalities of thought. New server architectures, UPSs on servers, battery plant grows as usage grows (including amortization). Optical and transponder costs: how do we squeeze that out? Client side needs to be cheaper, cost differential is too large. Router costs on backbone side is small. Dan: for us, leasing still makes sense. We’re not there yet in terms of transport.
Geographical markers that are challenging? Raj: Each region has uniqueness. India: cost structures of networking, maturity of networks, high capacity applications are challenges. Dan: there are certain regions that are a challenge, that prohibit us from exploring full potential.
Question about sustainability for green data centers? Do you factor this in when considering new centers and site choices? Raj: sustainability is one of top three criteria for new sites. They have a dedicated person on team that drives this. DOE grant on partnering in this area, is a very important consideration.
Question on do you get what you need from your suppliers (like carrier networks), or do you need to build new kinds or functionality in networks? Dan: as we focus on performance, we may have problems if network can’t keep up with certain levels of latency or service provider outages.
News Feed