Hardware Reference
In-Depth Information
6.10 [Discussion/15] <6.4, 6.5> Discuss some options to beter utilize the excess servers during
the off-peak hours or options to save costs. Given the interactive nature of WSCs, what are
some of the challenges to aggressively reducing power usage?
6.11 [Discussion/25] <6.4, 6.6> Propose one possible way to improve TCO by focusing on re-
ducing server power. What are the challenges to evaluating your proposal? Estimate the
TCO improvements based on your proposal. What are advantages and drawbacks?
Exercises
6.12 [10/10/10] <6.1> One of the important enablers of WSC is ample request-level parallelism,
in contrast to instruction or thread-level parallelism. This question explores the implication
of different types of parallelism on computer architecture and system design.
a. [10] <6.1> Discuss scenarios where improving the instruction- or thread-level parallel-
ism would provide greater benefits than achievable through request-level parallelism.
b. [10] <6.1> What are the software design implications of increasing request-level paral-
lelism?
c. [10] <6.1> What are potential drawbacks of increasing request-level parallelism?
6.13 [Discussion/15/15] <6.2> When a cloud computing service provider receives jobs consist-
ing of multiple Virtual Machines (VMs) (e.g., a MapReduce job), many scheduling options
exist. The VMs can be scheduled in a round-robin manner to spread across all available
processors and servers or they can be consolidated to use as few processors as possible.
Using these scheduling options, if a job with 24 VMs was submited and 30 processors were
available in the cloud (each able to run up to 3 VMs), round-robin would use 24 processors,
while consolidated scheduling would use 8 processors. The scheduler can also find avail-
able processor cores at different scopes: socket, server, rack, and an array of racks.
a. [Discussion] <6.2> Assuming that the submited jobs are all compute-heavy work-
loads, possibly with different memory bandwidth requirements, what are the pros
and cons of round-robin versus consolidated scheduling in terms of power and cool-
ing costs, performance, and reliability?
b. [15] <6.2> Assuming that the submited jobs are all I/O-heavy workloads, what are the
pros and cons of round-robin versus consolidated scheduling, at different scopes?
c. [15] <6.2> Assuming that the submited jobs are network-heavy workloads, what are
the pros and cons of round-robin versus consolidated scheduling, at different scopes?
6.14 [15/15/10/10] <6.2, 6.3> MapReduce enables large amounts of parallelism by having data-
independent tasks run on multiple nodes, often using commodity hardware; however,
there are limits to the level of parallelism. For example, for redundancy, MapReduce will
write data blocks to multiple nodes, consuming disk and potentially network bandwidth.
Assume a total dataset size of 300 GB, a network bandwidth of 1 Gb/sec, a 10 sec/GB map
rate, and a 20 sec/GB reduce rate. Also assume that 30% of the data must be read from re-
mote nodes, and each output ile is writen to two other nodes for redundancy. Use
Figure
6.6
for all other parameters.
a. [15] <6.2, 6.3> Assume that all nodes are in the same rack. What is the expected
runtime with 5 nodes? 10 nodes? 100 nodes? 1000 nodes? Discuss the botlenecks at
each node size.
Search WWH ::
Custom Search