Transcript SFT - CERN Indico

From Application to Appliance
Predrag Buncic
Overview
• Virtual Machines
• Appliances
 Hardware, software & virtual software appliance
• Demo
• Technology Forecast
 Clouds over Grid
• Conclusions
CERN, 30/03/ 2007 - 2
Back to the future…
• Some time ago…
 We had an application
• statically linked
• running in a VM
• prefect isolation
 Strict application boundary
• Then, things changed..
 Unix workstations, PC, commodity
computing
 Memory was still expensive
• Shared libraries, dynamical linking,
plugins
 Fuzzy application boundary
IBM-VM 360 mainframe, 1988
• Today
 Memory and disk space is cheap
 Virtual Machines running on commodity hardware on Open Source OS
are promising to deliver what we lost some time ago
CERN, 30/03/ 2007 - 3
Why (again) Virtual Machines?
• The system infrastructure can evolve independently from the
evolution of the application
• Re-creating application boundary
 Now we can Start, Stop, Pause, Migrate VM
 Software running inside a VM can not negatively affect the
execution of another VM
• VMs can provide a perfect process and file sandboxing
• The application can (re)use a lot of code which was previously
is system/kernel domain
 Fancy user space file systems via kernel modules
CERN, 30/03/ 2007 - 4
Virtual Appliances
•
This talk is about VM running your favourite application on your laptop
or desktop
Virtual Software Appliance = Application + Virtual Machine + Simple UI
•
Virtual Software Appliance is a lightweight Virtual Machine image that
combines


•
These appliances are designed to run under one or more of the
various virtualization technologies, such as

•
minimal operating environment
specialized application functionality
VMware , Xen, Parallels, Microsoft Virtual PC, QEMU, User mode Linux,
CoLinux, Virtual Iron…
Virtual Software Appliances also aim to eliminate the issues related to
deployment in a traditional server environment


complex configuration
maintenance
CERN, 30/03/ 2007 - 5
rPath: Software Appliance Company
CERN, 30/03/ 2007 - 6
rPath Technologies
•
Conary


•
rPath Linux

•
rPath's tool for building appliances
eliminates unnecessary components and provides only the software
needed by the applications on your appliance.
rPath Appliance Platform (rAP)

•
Linux distribution created as the basic operating system for Conarybased appliances
rBuilder


•
the underlying package management technology for rPath Linux
rPath provides instructions to package software from various popular
sources and package technologies
The rPath Appliance Platform (rAP) is rPath's extensible tool providing
a web-based user interface for maintaining appliances
rMake

rMake is rPath's tool for allowing full clean rebuilds of the software
used in an appliance when significant modifications are made to the
core operating system components
CERN, 30/03/ 2007 - 7
Practical exercise: AliEn Appliance
+
+
=
AliEn
External
Dependencies
busybox
ggbox
(system tools)
System devices
Kernel
Grid Appliance
CERN, 30/03/ 2007 - 8
Coda File System
•
CODA is a distributed file system with
its origin in AFS2. It is freely available
under a liberal license has several
features not found elsewhere:
•
disconnected operation for mobile
computing
high performance through client side
persistent caching and server replication
security model for authentication,
encryption and access control
continued operation during partial
network failures in server network
network bandwidth adaptation
well defined semantics of sharing, even in
the presence of network failures
•
•
•
•
•
/coda -> /opt/alien, /opt/packages
CERN, 30/03/ 2007 - 9
AliEnX
• AliEn Linux – minimal guest OS capable of running AliEn services
and hosting Grid applications
 http://alien.cern.ch/twiki/bin/view/AliEnX
 http://alien.rpath.org
• Built using rPath tools (rBuilder and Conary package manager)
 Very similar to what we were trying to do with AliEn BITS
CERN, 30/03/ 2007 - 10
AliEn Appliance
•
AliEn Appliance Version 0.4

x86 Mountable Filesystem (Xen Virtual
Appliance)
 x86_64 Mountable Filesystem (Xen Virtual
Appliance)
 x86 VMware (R) ESX Server Virtual Appliance
 x86 Installable CD/DVD
 x86_64 Parallels, QEMU (Raw Hard Disk)
 x86 Parallels, QEMU (Raw Hard Disk)
•
Already usable as User Interface


Generic, can be customized for other
purposes
To do:
•
•
•
Run Grid Jobs in
VM
Prototyping together with Globus developers
at Teraport cluster at University of Chicago

Start and manage VMs using Globus
Workspace Service
CERN, 30/03/ 2007 - 11
Benchmarks
• Test machine
 1 GB RAM
 3 GHz Pentium D
 100 particles, standard AliRoot setup
Xen 3.0.3
Native
Simulation
193 s
191.5 s
Reconstruction
52 s
51 s
CERN, 30/03/ 2007 - 12
Download history
CERN, 30/03/ 2007 - 13
Demo time…
CERN, 30/03/ 2007 - 14
AliEn Way: Grid Layered Cake
•
Large “physical grid(s)”

•
Individual V.O. have at given point in
time access to a subset of physical
resources


•
Reliably execute jobs, store, retrieve and
move files
Using standard tools to submit jobs (Job
Agents) to physical grid layer, V.O.
creates ‘upper’ middleware layer, an
overlay grid tailored to V.O needs but on
smaller scale
V.O has identity and can handle
interactions with physical layer on users
behalf
Virtual Cluster (User layer)
Virtual Grid (V.O. layer)
Physical Grid (Common layer)
Individual users interacting with V.O
middleware will typically see a subset of
the resources available to the entire VO

Each session will have certain number of
resources allocated
Reducing scale to achieve
scalability
CERN, 30/03/ 2007 - 15
Cloud over Grid
Physical
SQS
+ Grid
S3(Common)
+ EC2
•
Simple Queue Service (SQS) offers a reliable, highly scalable hosted queue for
storing messages as they travel between computers
• Simple Storage Service (S3) provides a simple web services interface that can
be used to store and retrieve any amount of data, at any time, from anywhere
on the web
 Elastic Computing Cloud (EC2) is a web service that provides re-sizable
compute capacity
CERN, 30/03/ 2007 - 16
rPath + Amazon
1. Software developers use rBuilder to build an Amazon
2.
•
Machine Image (AMI) that is stored using the Amazon
Simple Storage Service (Amazon S3).
With a single click, rBuilder and rBuilder Online users can
boot their software appliances on Amazon EC2.
User has a complete control of instances




http://www.rpath.com/amazon
•
1.7Ghz x86 processor
1.75GB of RAM
60GB of local disk
250Mb/s of network bandwidth.
User can
•
•
•
load them with their unique software appliance image
manage their network's access permissions
and run their image using as many or few systems as they
desire.
CERN, 30/03/ 2007 - 17
$$$ vs ???
•
Everything has the price…



•
$0.15 per GB-Month of storage used
$0.10 per instance-hour consumed
$0.20 per GB of data transferred in/out
How much would it cost?


$
?
Store 10PB/year
Use 40k CPUs for processing
storage
1.8
10000000
$18,000,000.00
in
0.2
10000000
$2,000,000.00
out
0.2
10000000
$2,000,000.00
cpu
0.1 350400000
$35,040,000.00
•
•
•
•
We have faster network
We have 40k CPUs
We have the storage
Why don’t we have 99.99%
service availability?
$57,040,000.00
CERN, 30/03/ 2007 - 18
What did they do differently?
•
Storage

Scale as an advantage
•
Adding nodes to the system increases, not decreases, its availability, speed,
throughput, capacity, and robustness
• P2P model

99.99% data availability
•
•
•
Computing

Elastic
•



EC2 enables you to increase or decrease capacity within minutes, not hours
or days
Integrated
•
Designed for use with S3
Secure
Uniform
•
•
There can be no single points of failure
Redundant data copies on spinning media
Uses VM technology to provision uniform representation of computational
resources to end user
Security and messaging



Web (portal) and Web services provide entry points into the system
But internally, system uses reliable message queues
Event driven rather than service oriented implementation
CERN, 30/03/ 2007 - 19
Use cases for Virtual Machines
•
Grid


•
VO box

•
Enhanced Scalability
User Interfaces


•
Sandbox environment for job execution on WN
Enhanced site security
Separation of Grid and system environment
Reducing Grid initiation threshold
Specialized environments

PROOF/CAF
•
•
•
•
Training setups

•
process migration
kernel modules to enable fancy user space file systems
P2P like object sharing and caching
Make sure that everyone has the same environment when they walk in
training room
Testing environments

Easy to setup, saving time and money
CERN, 30/03/ 2007 - 20
Conclusions
•
Virtual Machine are coming back as viable technology with lots of
potential benefits for users and resource providers
•
The technology is maturing quickly


•
If we do not want to end up paying a big money to big business we
have to learn from them just as they learned from us

•
lots of technology behind the clouds is the same Open Source software
under our fingertips
Are VMs going to solve all our problems?

•
The business is catching up fast…
… and overtaking us even faster
No.
Do they have a potential to make our life simpler and computing less
expensive?

Yes.
CERN, 30/03/ 2007 - 21