Distributed, Robust Auto-Scaling Policies for Power Management in Compute Intensive Server Farms

Abstract

Abstract-Server farms today often over-provision resources to handle peak demand, resulting in an excessive waste of power. Ideally, server farm capacity should be dynamically adjusted based on the incoming demand. However, the unpredictable and time-varying nature of customer demands makes it very difficult to efficiently scale capacity in server farms. The problem is further exacerbated by the large setup time needed to increase capacity, which can adversely impact response times as well as utilize additional power.In this paper, we present the design and implementation of a class of Distributed and Robust Auto-Scaling policies (DRAS policies), for power management in compute intensive server farms. Results indicate that the DRAS policies dynamically adjust server farm capacity without requiring any prediction of the future load, or any feedback control. Implementation results on a 21 server test-bed show that the DRAS policies provide near-optimal response time while lowering power consumption by about 30% when compared to static provisioning policies that employ a fixed number of servers.

Deadline-Aware Energy Management in Data Centers

“…The methods aiming to reduce data center power consumption could be classified into four approaches [1]: powerproportionality, which attempt to guarantee that servers consuming power in proportion to their utilization [2]- [4]; energyefficient server design, which attempt to determine the proper server architecture for a given workload [5]- [7]; dynamic server provisioning, which attempts to determine the times when the servers should be kept on or off [8], [9]; consolidation and virtualization, which attempts to reduce the power consumption by resource sharing [10]- [12]. We refer the reader to [13]- [17] for further literature related to our work.…”

Heterogeneity and thermal aware adaptive heuristics for energy efficient consolidation of virtual machines in infrastructure clouds

“…Power state transitioning of servers' processor involves a transition time latency or time overhead, which is highly processor dependent and this could vary from one processor type to another ( Table 4). The server stays in an interim power state called SETUP state for the period between each such transition and consumes power called SETUP power [2].…”

“…Interim server processor state during the state transition is referred to as SETUP power state. Power drawn during wake-up and shutdown transition time period can be considered equal to the server power consumption at highest utilization load [2]. To improve datacenters' power efficiency, servers are to be well utilized and when not in use are to be switched OFF or transitioned to low power SLEEP states, hence the importance of server consolidation and processor power state transition management.…”

Tales of the Tail

“…In DRAS, the authors explored policies for load balancing requests across servers and deciding when to power idle servers off, assuming computationally intensive workloads [10], demonstrating that a reduction in average power can be achieved with a slight penalty in average latency. Similar observations were made for the Salsa web server [8], though they also demonstrated that request batching could further save energy.…”