Unraveling Multimodality with Large Language Models.pdf
Cloud Computing: An Introduction
1. Cloud Computing
and Virtualization-
Newest Trends
Srinath Perera, Ph.D.
Senior Software Architect, WSO2
Research Scientist, Lanka Software Foundation
Member, Apache Software Foundation
Visiting Faculty, University of Moratuwa
2. Next 1 hour: Goal
What is Cloud?
More on Why and
When and less about
How?
Take home is some
idea of when to and
when not to use the
Cloud
3. Men, it has been well said, think in herds; it will
be seen that they go mad in herds, while they
only recover their senses slowly, and one by
one. ~ Charles Mackay
Copied from http://www.flickr.com/photos/54555810@N00/2848637190/, by Rambling Traveler
4. Gold Rush
Very good example of a hype
Only few actually made money
But associated services (merchants
and transportations) made lot of money
5. The Cloud Bandwagon
Is Cloud a hype? Of course
it is!
Is it Just hype? may be not,
thats what we will discuss.
But don’t get me wrong,
even if it is a hype, that
doesn’t mean we should
not be talking about it.
Image from http://www.flickr.com/photos/88929764@N00/4436978855/
6. Some Hypes Deliver: Hype Cycle
We have to understand what it is and what
drives it? That would tell us how to navigate.
7. Selling Shovels In the Gold Rush
Hype change the demand
and supply dynamics.
8. Fire Open up the space for new
Trees
Hypes often change the dynamics, and give a
chance for new comers to climb to the top
Image from http://www.flickr.com/photos/leppre/485528374/
9. So What is Cloud?
Based on the idea that computation and storage
can be rented as a utility from data centers that
runs somewhere (in the cloud) on demand.
− Remote resources that are rented
− On demand and in elastic manner
− Pay as you go
10. Electricity as a Utility as an
Example
Now, no one run generators themselves
Use electricity that is remotely generated
Can draw when need it
Only pay for what you use
11. Electricity as a Utility: Benefits
Small startup cost (do not have to buy
a generator)
No operational cost
Do not need to do capacity planning
Overall cheaper electricity due to
economics of scale
Making it parts of the everyday life,
commodity (accessibility)
Image from http://www.flickr.com/photos/infomofo/3220498521
12. Benefits of the Cloud
Avoid/reduce operational
costs by outsourcing
Can scale up and down as
needed
Pay as you go
Making it parts of the
everyday life, commodity
(accessibility)
Cheaper computing power
due to economy of scale
13. Drivers of Cloud
Unused computing power at Google, Amazon
Max load >> average load
High operational cost, need for outsourcing
Availability of large scale solutions and
infrastructure as side efforts of high tech
company operations
15. IaaS Example: Amazon
Two services
− S3 Storage
− EC2 computing cloud
Based on Virtualization, where each user is given a
virtual machine and charged by the hour
Need least amount of changes to move apps to the
Cloud. They in a way replace hosting services
Least amount of out of the box services (e.g. DOS
attack prevention) and advanced services like scaling
etc., are a responsibility of the user.
Often the best choice for ad-hoc computer users.
16. PaaS Example: Google App
Engine
Support Java and Python
Support web requests and run user written web
applications in an isolated environment
− Java version based on servlets
Support storage based on Big table, memcache
based cache, and auto scaling
Can write apps locally, test it, and then upload
to the Cloud
17. SaaS Example: Salesforce
Provide support for CRM (Customer
Relationship Management) software as a
Service
The application available out of the box users
just configure and use it.
Salesforce handles all the details, and a ideal
choice for outsourcing IT functions
However, applications are very specific and
customizations are limited.
18. Virtualization
Mimic the hardware layer using software, and
provide a “virtual” machine to the next layers
− Used to mimic multiple machines
within the same machine
− easier manageability and migration
(move machines based on load)
− e.g. Vmware, Xen, KVM
IaaS uses Virtualization to provide infrastructure as a
service
Virtualization can add significant overhead (each
instruction become 2 instructions)
− new CPUs have hardware support for virtualization,
which make things better
− Still I/O is a challenge
20. Private Cloud
Run a Cloud within the
organization (mainly due
to security concerns).
e.g. Amazon Virtual Private Cloud
(VPC), IBM private cloud, WSO2
Stratos private Cloud
Idea is optimizing resource sharing, utilizations,
and operations
− e.g. testing environments
Connection to public Cloud is possible
− Amazon VPC uses VPN, WSO2 Cloud Service
Gateway
Image http://www.flickr.com/photos/glasgows/536185797/, Some rights reserved by M Glasgow
22. More Room to Outsource Non
Key Functions of a Organization
Organizations outsource their
non-competitive areas to
reduce costs and focus on their
own expertise.
IT departments are a major
cost in most organizations
Cloud enables Organizations to
outsource some of the IT
functions
23. Small Start Up Cost
Small start up cost
− Reducing the gap
between visionaries and
dreamers
− New organization has
better chance for
outsourcing operations
through the Cloud
− Cloud competition likely
to drive down hosting
prizes
24. 1 computer for 100days = 100
computers for 1 day
Great tool for occasional
computations
− Research labs
− Reporting collecting data
for a story
− New york Times tiff to pdf
conversion
Rise of analytics
− great tool for offline
analysis
− Business Intelligence (BI)
−
25. Large Scale Computation and
Storage Resources becoming a
Commodity
Normal people, not
just organizations
can have access to
computing power
and storage
Computer intensive desktop apps (e.g. Excel, 3D Max)
26. Autoscaling and Cloud Bursting
Max load >> average load
Allocate based on the load
Running internal machines in an average load
(because cloud is still expensive than hardware ) and
scale out to cloud when there is high load
− Mimic the Hybrid car
28. Privacy/ Security/ Trust
With cloud you will run your apps and put your
data in an outsider's administrative domain
− Can we trust the outsiders to not
look at our data? Well it depends.
But concern is normal for any out
sourcing
− Can we trust their security
measures? Does the isolations
are good enough?
− If we are keeping data owned by
others, what are the legal
implications etc.
29. Latency/ Bandwidth
Why does electricity as a Utility was so
successful? One key aspect was almost
unlimited speed and capacity
Is that the same for Cloud?
− Not really, specially for countries like
Sri Lanka
− Most people dismiss this, and does
not even want to discuss
− But for some apps (e.g. interactive
apps like games) this can be a issue
Possibilities
− support for offline operation
− Fedex your data
30. Performance
In the Cloud, your
software will run on an
another layer of
abstractions
It will inevitably slower (3-4 times if unlucky)
− Often the overhead comes from I/O
− Some hit on CPU power
− Expect the bottlenecks to shift
Remember performance ≠ scalability
− Cloud likely to let you scale out, but performance
on individual nodes likely to go down
31. Challenges
Cloud Middleware
− Long running, large scale, fault Tolerant
Computations
− Scalable service and resource
scheduling
− Scalable, secure, self-managing, fault-
tolerant data storage for long running
processors
− Multi-tenancy and Isolation
− Delegation across organizations
Cloud Operations
− Metering and Billing
− Provisioning and Monitoring
− Data Confidentiality and auditability
36. Look back: Recommendations
(Contd.)
If your services have a Max load >> Average load, you
should think about crowd bursting
If you do heavy computations once in a while (analytics,
audits), then use IaaS there.
If you want to outsource some of your IT functions, think
SaaS
Before leap think about 3 major concerns of Cloud?
− Security/ Privacy, Latency and Performance
How much sharing and savings I need?
− IaaS, PaaS, SaaS