NSF CSR: Using Burstable Instances For Cost-Effective Tenant Orchestration

in the Public Cloud

B. Urgaonkar (PI) and G. Kesidis (co-PI)

 

            2017-2020           

           

Public cloud computing providers are offering myriad virtual machine (instance) types that, in different ways, trade off lower prices for poorer capacity. Whereas cheaper revocable instances, e.g., spot instances from Amazon EC2 spot or preemptible instances from Google Compute Engine (GCE), have been studied quite extensively, a recently emergent alternative type - "burstable" instance - has received scant attention. This proposal argues that burstable instances, currently offered by EC2 and GCE, present many tenants with significant opportunities for cost- and performance-effective operation. To realize this potential, however, tenants must develop a thorough understanding of how these instance types operate - something lacking in the state-of-the-art. The goal of this proposal is to develop such an understanding. This proposal identifies key idiosyncrasies of capacity dynamism for burstable (and other low cost, “wimpy”) instances that sets them apart.

           

 

         Papers and personnel participating in work supported in whole or part by this grant:

 

·      J. Khamse-Ashari, I. Lambadaris, G. Kesidis, B. Urgaonkar, Y. Zhao. A Cost-efficient and Fair Multi-Resource Allocation Mechanism for Heterogeneous Servers. In Proc. IEEE GLOBECOM, Abu Dhabi, Dec. 2018.

·      Y. Shan, A. Jain, G. Kesidis, B. Urgaonkar, J. Khamse-Ashari, I. Lambadaris. Scheduling distributed resources in heterogeneous private clouds. In Proc. IEEE MASCOTS, Milwaukee, Sept. 2018.

·      J. Khamse-Ashari, G. Kesidis, I. Lambadaris, B. Urgaonkar and Y. Zhao. An Efficient and Fair Multi-Resource Allocation Mechanism for Heterogeneous Servers. IEEE TPDS, May 2018.

 

Software prototypes:

·      https://github.com/yuquanshan/spark/tree/d-vector

·      https://github.com/yuquanshan/mesos/tree/multi- scheduler

 

Some earlier papers on related work:

 

·      C. Wang, B. Urgaonkar, N. Nasiriani, and G. Kesidis. Using Burstable Instances in the Public Cloud: When and How? In Proc. ACM SIGMETRICS, Champaign-Urbana, IL, June 2017.

·      Ibid. Errata.