Quiz 3, Feb 2, 2011
CSE 591 Topics in Green Computing and Communication, Spring 2011
Time: 10min Points: 10
1. (1 pt) If a datacenter pays $ A for each amp usage per year, what would be the total annual
electricity bill for using N severs consuming W watts each. Assume V Volts power supply and PDUs
have E% efficiency.
2. ( 1 pt)Assume an N core chip and suppose we can Turbo Boost one core X times. For a given
computation what speedup you can get if you can use all N cores (parallel execution) without Turbo
Boosting as opposed to Turbo boosted single core performance (turbo boosted sequential
3. (2 pt)Now suppose the application has s% sequential code (which can be turbo boosted) and parallel
part can utilize all N cores (with no turbo boosting). What speedup (against non-parallelized non-
turbo boosted code) can be obtained for this application?
XN/(sN + X(1-s))
4. (2 pt)Suppose for three runs of an optimized code we get speedups of 2, 12, and 1/3, respectively.
What is the arithmetic mean and what is the geometric mean of the speedups?
AM = (2+12+1/3)/3 ; GM = (2*12*1/3)^(1/3)
5. (1 pt)What is the difference between “balanced power plan” and “high-performance power plan”.
High perf allows turbo boosting
6. ( 1 pt) What are the advantages of running a CPU with a guaranteed thermal limit in a datacenter?
Limits the maximum current drawn. Current usage beyond a limit costs heavily for DCs.
Won’t exceed cooling capacity. Reliability benefits. good usage of space.
7. ( 1 pt) Why does a benchmark simulating a High Performance Computing situation not applicable for
a data center?
HPC datacenters usually run at higher utilization (say 50 to 90%) than typical data centers (30 to
60%). Also these workloads may be compute intensive as opposed to I/O intensive for typical data
8. ( 1 pt) Why do system administrators of data centers try to avoid letting CPU utilization rise above
70% where CPUs may be most efficient?
Usually SLA are based on response time. At high utilization the response time increases
exponentially leading to frequent SLA violations.