Thoughts on Open Cirrus Cloud Computing Testbed: Federated Data Centers for Open Source Systems and Services Research

Authors: Roy Campbell,5 Indranil Gupta,5 Michael Heath,5 Steven Y. Ko,5 Michael Kozuch,3 Marcel Kunze,4 Thomas Kwan,6 Kevin Lai,1 Hing Yan Lee,2 Martha Lyons,1 Dejan Milojicic,1 David O’Hallaron,3 and Yeng Chai Soh2

1HP Labs, 2IDA, 3Intel Research, 4KIT, 5UIUC, and 6Yahoo!

Venue: HotCloud '09

Paper: http://www.usenix.org/events/hotcloud09/tech/full_papers/campbell.pdf

Summary:

The authors present the need for systems research on datacenter infrastructure. While virtualized systems like EC2/S3, Azure, and such allows applications research, it is difficult to do research on systems that compose the infrastructure of the data center. Open Cirrus is an effort to provide researchers with access to hardware level end-hosts. They have fedarated several data centers around the world, and are working on providing a unified API to access all of them. There will be some basic common functionality even though the DCs themselves are heterogeneous. OpenCirrus is basically federated Emulab across multiple sites where the sites are only loosely coupled (whatever that means).

The systems are scheduled and manage as Physical Resource Sets (PRS). HP's integrated Lights-Out technology is used to manage machines at the firmware level.

They compare options for testbeds and do a back-of-the-envelope calculation of the break-even point for renting (S3/EC2) vs. owning hardware (they use some reasonable cost-breakdown). There calculation says that for a service running more than 12 months, it is better to buy computation while for storage more than 6 months it is better to buy.

The paper is interesting in that it surveys other testbeds, and introduces a new one wider scale one than emulab.