OxGrid: Virtualisation at Oxford Oxford eScience Rhys Newman

advertisement
Oxford eScience
OxGrid:
Virtualisation at Oxford
Rhys Newman
Manager of Interdisciplinary Grid Development, Oxford University
Campus Grid Workshop – Edinburgh June 2005
Spare Computer Resources

Yesterday’s analysis shows how the costs of a
campus grid will be about £30k per week.
• Still much cheaper than £6 million to buy the machines
to do the same work, and still have to pay for the
electricity!

Assuming 20000 machines available with average
age less than 3 years, there may be up to 500TB
available
• Still cannot quantify this figure precisely.
• But storage is easier to manage with simpler technology,
and does not have the associated running costs.
Oxgrid Client


A background “screen saver” to manage
idle/spare resources.
Developed in Java for cross platform properties
• A very useful attribute for heterogeneous computing
environments.

Initially will be used to assess how much CPU
time and storage can be harnessed without
impacting users (machine owners).
Oxgrid Client



Most computers (90% or more) run Windows, so
initial prototype designed for that, rather than
YALA.
Oxgrid Client runs in the background and can be
configured by the Oxgrid Icon in the system tray.
End user (host machine) acceptance:
• Does it interfere with normal work?
• Does it cause any instability of the machine?
• Is it easy to use and understand (provider’s
perspective)?
Oxgrid Client
Oxgrid Client
Advanced Schedule
Emulation / Virtualisation


A Virtual Machine can be frozen on
command to provide OS level
checkpointing!
A management system can monitor
progress of jobs and continually match
them to idle resources, moving them at
will.
Grid Job
Guest OS
Emulated hardware
other
software
Emulator
host OS
host hardware
Software Only:
1. Storable/Freezable
2. Relocatable over the
network
Fundamentals



Moving a whole disk image of an OS
around the network to find a CPU.
Storing whole disk images until free
processing is available.
But technology is changing the design
considerations.
• CPU x2 in 18 mths, NW x2 in 9 mths, Storage
£/MB halving in 6 mths.

Systems desiged in the past are built
around different assumptions – not
necessarily the best suited in this new
world.
OxGrid: Operation
OxGrid Operation Centre
Researcher:
needs lots of computer time/storage
Manages idle/spare resources
“100 jobs in 24 hours please”
Various Idle
Resources
Running OxGrid Client
First 2 hours here
Next 10 hours here
6 Hours here
Final 4 hours here
Need to finish in 24 hours
Each job may only take 12 hours
OxGrid turns many unreliable resources into one large reliable resource
OxGrid


Central Management of virtualised
resources around Oxford University.
All key technologies and equipment
needed already exists:
•
•
•
•

Networking, internet technologies.
Virtualisation.
Idle machines and spare storage.
real users with real needs.
Activity does not fit into an existing
department:
• too much
computer
• too much
computer
like product development for
science
like product development for
services.
Virtualisation as the Enabling
Technology



Virtualisation enables existing resources to
be leveraged without complex security and
software engineering constraints on users.
Virtualisation guarantees resource
providers security and acceptable use by
3rd parties (owner’s perspective).
Core technology is open source, technique
is industry standard and no longer suffers
from performance issues:
• 10 machines at 80% is better than 1 at 100%!
Interesting Opportunities
+





=
With virtualisation computing becomes a
liquid commodity….
A commodity can be traded…..
A traded commodity can have derivative
instruments…..
Value can be derived for the main and
derivative markets.
What commercial interests could be here
either as customers or providers?
Coming soon….
If Oxgrid can use idle resources
so well within Oxford,
then it should be as easy to share
machines between Universities……
Download