Cloud Computing
Francis C.M. Lau, HKU
“The sun always shines above the clouds.”
- Paul F. Davis
Big Data and Cloud
• We embrace cloud not just because we need to
process data
• Also because we need a platform (PaaS), certain
software (SaaS), or hardware resources (IaaS)
• But true, Big Data made cloud happen a lot
more quickly
– You don’t want to operate a power plant at home just to
control a power-thirsty appliance
2
Cloud as Utility
“The long dreamed vision of computing as a utility is
finally emerging.” [Armbrust et al.]
• You plug in (the outlet) and play [but sometimes it won’t]
• You thought it is an infinite power source [but sometimes it’d
run low, or even run out; and more often, it behaves unstably]
• You assume it is “elastic” – you use what you need exactly and
pay for just that [but sometimes it won’t stretch, sometimes it
breaks, and you’re charged unfairly]
• You thought everything is pretty safe [but didn’t realize it
could be a black hole]
3
Subtopic: Service Availability
• Dropbox “dropped out” on Jan. 10, 2014 for 2 days
• Clouds are a huge assemblage of components, and
software has bugs!
• If your server at home hangs, you reboot, but you can’t
when a cloud hangs
• Distributed Denial of Service (DDoS) attacks are real
• RQ: How to design a cloud service that is highly
available?
• RQ: How to counter the “attacks”?
• RQ: How to tolerate faults or failures of
components?
4
Subtopic: Performance
Predictability
• Fact: most virtualized environments have highly variable
performance
• Variance also due to multi-tenancy, movements of large
amounts of data, and the system itself (e.g., HDFS
randomly distributes data blocks across a cluster)
• Even if CPU and memory sharing is not a problem, I/O
sharing could easily kill performance
• Many HPC applications need to ensure that all the
threads of a program are running simultaneously
• RQ: How to make performance more predictable?
• RQ: How to guarantee performance/QoS?
5
Subtopic: Providing Elasticity
• Scalability is key: quick, automatic scale up or down according
to user’s changing needs
• Application’s scalability is another issue
– 1 machine x 100 hrs = 100 machines x 1 hr?
• Ideally, you pay as you go, and are charged by the cycles
(compute), or the bytes (storage and communication)
• RQ: How to predict and react to workload changes
quickly and dynamically?
• RQ: How to reduce bottlenecks and provide for the
best speedups?
• RQ: How to charge more accurately and fairly?
• RQ: How to scale data storages?
6
Subtopic: Data Confidentiality
“The main issue is that expectations of
trustworthiness may be unrealistic.” [Neumann]
• Apparently there should be no “fundamental” obstacles
to making a cloud-computing environment as secure as
in-house IT environments
– But clouds do have a lot more weak spots
• Gartner: 50% of enterprises will use hybrid cloud (which
includes a private cloud) by 2017
– Also for performance reasons: some data are “earthly”
• RQ: How to make cloud sufficiently secure and
trustworthy?
7
Subtopic: Data Lock-In
• Although software stacks have improved
interoperability among platforms,APIs for cloud
applications are still predominantly proprietary
• Customers cannot easily extract their data and
programs from one site to run on another
• It is really “vendor lock-in”
• RQ: Standardization of APIs?
• RQ: How to design a heterogeneous cloud
that would integrate parts from multiple
vendors?
8
Subtopic: Optimizing Data
Placement and Transfer
• Big Data: applications easily get “pulled apart” across
the boundaries of machines or even clouds
• Cost and performance depend a lot on data
placement and transport
– Jim Gray:The cheapest way to send a lot of data is to
physically send disks or even whole computers via
overnight delivery services
• RQ: How to place and re-place data such that
the best cost-performance can be achieved?
9
“The ICT ecosystem (the Internet,
Big Data, and the Cloud) now
approaches 10% of world
electricity generation”
• Amazon: energy-related costs: 42% of total (19% power; 23% cooling)
[2009] (now much improved)
• Cloud computing (due to server consolidation) is considered green
computing, but the computers they use may not be green
10
Subtopic: Green Cloud
• Existing solutions: Energy efficient hardware,
processor-level energy-aware scheduling (e.g., DVS)
• Even when run at a low utilization, servers typically
need up to 70% of their maximum power
consumption
• Virtualization increases energy efficiency
• RQ: How to perform energy-aware scheduling?
• RQ: How to achieve the best tradeoff in
computation/communication/storage and
energy/performance?
11
Emerging Opportunities
• Thin interactive apps that are backed by the cloud,
even when they are disconnected
– Mobile cloud
– Edge computing, fog computing
• Cloud and IoT
– Most “things” are not computers
• Data intensive batch processing for business analytics
– Less online transactions, more decision support
• Compute-intensive desktop apps
– Symbolic math, 3D rendering, …
12
More RQs by Colleagues
• Cloud accesses are remote and have low performance. Caching
improves performance but is subject to reliability challenges. How
to design high-performance and high-persistent caching strategies?
• Integrating multiple clouds (cloud-of-clouds) can boost scalability,
but how to address the heterogeneity of different clouds?
• How to design dynamic pricing mechanisms that are optimal?
• How to support online education and remote health through a
cloud platform?
• How to jointly optimize network and data resources in order to
achieve effective geo-diversity in datacenter design?
13
… Hong Kong
• Ideal location for datacenters, data hub
– Cf. the “Enhancing Hong Kong's strategic position as a regional
and international business center” theme
• Green cloud
– Cf. the “Developing a sustainable environment” theme
• Mobile cloud
– HK ranks #1 by connections/citizen (March 2015)
• Adoption by SMEs and startups
– “It used to take years to grow a business to several million
customers – now it can happen in months.” [Armburst et al.]
• We’re very strong in Data Engineering, Networking,
Cloud, …
14
References
• Michael Armbrust et al.,“A View of Cloud Computing”, CACM,Volume 53
Issue 4,April 2010.
• Andreas Berl, Erol Gelenbe, Marco di Girolamo, Giovanni Giuliani,
Hermann de Meer, Minh Quan Dang, and Kostas Pentikousis,“Energy-
Efficient Cloud Computing”, The Computer Journal,Vol. 53 No. 7, 2010.
• Ken Birman, Gregory Chockler, and Robbert van Renesse,“Toward a cloud
computing research agenda”, ACM SIGACT News,Volume 40 Issue 2, June
2009.
• Peter G. Neumann,“Inside Risks Risks and Myths of Cloud Computing and
Cloud Storage”, CACM,Volume 57 Issue 10, October 2014.
• Malte Schwarzkopf, Derek G. Murray, and Steven Hand,“The Seven Deadly
Sins of Cloud Computing Research”, HotCloud 2012.
15
