Description
1. Introduction
You are hired by a startup company who is considering to use cloud computing instead of building its own
infrastructure.
There is concensus that a cloud computing software stack at the layer of IaaS will be used, but its not
clear whether the computing resources should be rented from a public cloud on-demand, or whether a private cloud
should be purchased.
You are tasked to find the cost breakdown of a private cloud, and compare that to what
Amazon would charge. You can find many instance types defined at http://aws.amazon.com/ec2/instance-types/,
and their prices are set at http://aws.amazon.com/ec2/pricing/. For pricing purposes, please stick to Linux ondemand pricing.
Since you have to estimate the cost of the hardware when building a private cloud, you can use hardware prices
found at Dell (http://www.dell.com/p/enterprise-products.aspx?c=ae&l=en&s=bsd&~ck=mn), AcmeMicro
(http://www.acmemicro.com), or Pogolinux (http://www.pogolinux.com) as good sources for server hardware.
If
you cannot find some particular hardware here, please cite whatever site you find where you obtained the pricing
information.
You must include a printout of your shoping cart in your final writeup report for this assignment; include
this as an appendix at the end of your report.
You are to estimate the cost of different configurations for 3 different set of requirements:
• Configuration 1: Hadoop/Spark Cluster with 32K-cores, 256TB memory, 50PB HDD, and 10Gb/s Ethernet
Fat-Tree network (each VM should be equivalent to the d2.8xlarge instance); in addition to the compute
resources, a 100PB distributed storage shared across the entire cloud should be procured, with enough
capacity for 100GB/sec throughput (for pricing comparison, see S3)
• Configuration 2: Support 1 million virtual machines (VM) where each VM requires 2-core, 15GB RAM, 32GB
SSD storage, and 1Gb/s Fat-Tree network (each VM should be equivalent to the r3.large instances); in
addition to the compute resources, a 10PB distributed storage shared across the entire cloud should be
procured, with enough capacity for 10GB/sec throughput (for pricing comparison, see S3)
• Configuration 3: Support deep learning with 1 exaflop of mixed precision performance (hint: each VM
should be equivalent to p3.16xlarge instances; you will want to use the NVIDIA V100 GPUs (8 GPUs per
node), and allocate 8-cores per GPU (64-cores per node) with 8GB of memory per core (512GB per node);
the network to use is at least 10Gb/s per GPU (100Gb/s should work), and should be organized in a Fat-Tree
network; in addition to the compute resources, a 1PB distributed storage shared across the entire cloud
should be procured, with enough capacity for 10GB/sec throughput (for pricing comparison, see S3)
2. What you will submit?
Your deliverables for this project are to be written in a report, which will include the following:
• Compare the costs of the 3 different configurations between the public cloud (Amazon AWS) and the private
cloud
o you may assume a 5 year amortization cost
o you will have to factor in things other than hardware, such as cooling, power, administration costs,
network infrastructure (e.g. switches); you can assume 1 system administrator is needed for every
1000 servers
o show your data in three different tables with the costs of each of the 3 configurations, broken
down by components (e.g. servers, network switches, cables, racks, cooling, power,
administration, etc)
o summarize your data in a 4th table, comparing the public cloud cost to the private cloud cost
• Explain in words if it is better to rent or buy. If it is better to buy, what utilization must you maintain over
the 5 year lifetime of the private cloud in order to break even on the investment?
Table 1: Table summarizing each of the three configurations; please include 1 table per configuration, for a total of 3 tables
Description Price per Item Quantity Total Price
Compute Servers
Network Switches
Network Cables
Racks
Storage Servers
Electric Power
Cooling
Administration
TOTAL N/A N/A N/A
Table 2: Summary table comparing the 3 configurations between the public and private cloud; your cost of power, cooling, and administration should be
to cover 5 years of costs
Configuration 1 Configuration 2 Configuration 3
Public Cloud (including EC2 and S3)
Cost over 5 years, 24/7 operation,
with 100% usage
Private Cloud cost over 5 years, 24/7
operation, with 100% usage
What utilization must be achieved
with the private cloud to make the
private cloud option more attractive
than the public cloud?
You are to make a PDF of your project writeup. At the end of your PDF, include a printout of your shoping cart
showing the hardware you would need to purchase. Submit a single PDF file on BB, with the following filename
“Proj_LASTNAME1_LASTNAME2.pdf”. Please ask questions on Piazza if there are any doubts about assumptions you
need to make.