At the Silicon Valley Cloud Computing Group: Zynga presents a Real World Case Study

If you believe cloud computing is still in its infancy, you have not learnt about Zynga. While Zynga have a hybrid cloud that powers their game servers involving their data center, a private cloud, zCloud, and a public cloud, namely Amazon Web Services (AWS), the more interesting fact is that they expect to be profitable in 2011 with a net margin of 35%, or a net profit of $630m on a $1.8B revenue.


At the Cloud Computing meetup today, Allan Leinwand, Zynga's CTO - Infrastructure, presented a fascinating study of how Zynga has employed cloud computing to power their game servers. It appears we will not see Allan's presentation in a publicly accessible web site any time soon — See Sebastian's talk about the meetup — and what follows is an account of the meeting, transcribed from my notes written out on paper. By the way, Allan's rapid fire delivery is a treat to experience.

Zynga started out renting server power through Amazon Web Services, and managing them by RightScale. By 2009, need for greater flexibility and scalability¹, tipped Zynga to build a private cloud, zCloud, and complement it by using AWS. To smoothen the hybridization effort, they built zCloud as an AWS look-alike, including their security model, etc. In building zCloud, Zynga also avoided the use of inter-rack VLANs, and used an IP-routed network to interconnect racks.

Some interesting details about zCloud:

1. Zynga uses 1 VM per physical server.
2. CentOS is the guest operating system.
3. They use x86-based physical servers.
4. They have created a highly automated scheme whereby they can provision > 1000 physical servers in < 24 hours.
5. The traffic in their cloud is more North-South, between tiers, than East-West (across servers).
6. Growth: In January 2011, the server count was 75 times that in January 2009².
7. A single orchestration layer manages the three key resources of the cloud: CPUs, Storage and Network.

Allan's wisdom to application designers: "Understand the architecture of the cloud in designing your application."

Allan's recommendation for "Cloud 2.0":

1. 99.999% availability at the IaaS level
2. Compute power needs to scale to millions of CPUs. (Can the cloud borrow concepts from SETI@home?)
3. The cloud environment needs to be able to handle any workload

Overall, a very enjoyable and highly educational evening. Box.net — another vendor powered by a cloud — were gracious to provide the venue for the evening and Sebastian & Co. (i.e., whoever else worked in the background to make this event happen) deserve a big round of applause.

---

¹In addition, the usual OpEx vs. CapEx argument was also a factor.

²For obvious reasons, the absolute server count was not revealed.