private cloud reduces risk calculations by 50%

© 2008 TIBCO Software Inc. All Rights Reserved. Confidential and Proprietary.

TIBCO (DataSynapse) Grid Server

David Broeren

Business Architect

NAB


2

Company Overview

National Australia Bank is among Australia’s largest

financial institutions by market capitalisation and one of

the world's top 30 financial services companies.

NAB operates across 10 countries serving 10 million

banking and wealth management customers.

Wholesale Banking, a division of NAB, has more than

2,500 people across Australia, New Zealand, Asia, the

United Kingdom and the United States.

Wholesale Banking provides customers with access to

international financial markets and a range of

specialised funding, investment, asset services and risk

management capabilities.http://www.wholesale.nabgroup.com/AboutUs/Pages/Whoweare.aspx

http://www.wholesale.nabgroup.com/AboutUs/Pages/Whoweare.aspx


3

Introduction: Setting the Context?

Server or

Container

Server or

ContainerHardware Virtualisation

Virtualisation

Container Container

Deployment Grid

Server or

Container

Server or

Container

App

2

App

1App

1

Grid (Server)

App

2

App

1

Grid (Desktop)

Client2Client1Desktop or

Citrix Svr

Client1

Cluster

Server or

Container

Server or

Container

Server or

Container

App

4

Data

Grid

Compute

Grid

Server or

Container

App

5Software as a Service

App

1

App

1

Virtual Server

What is Cloud Computing? What is a Grid?

http://images.google.com.au/imgres?imgurl=http://regmedia.co.uk/2006/11/01/core2extreme_quad_cpu.jpg&imgrefurl=http://www.reghardware.co.uk/2006/11/02/review_intel_core_2_extreme_qx6700/&h=411&w=450&sz=72&hl=en&start=4&tbnid=tzWHn57CUB_9dM:&tbnh=116&tbnw=127&prev=/images%3Fq%3DCPU%26gbv%3D2%26svnum%3D10%26hl%3Den


http://www.myfastpc.com/ram.jpg


http://images.google.com.au/imgres?imgurl=http://www.dbgweb.com/CompaqAlphaServerES45Pedestal.JPG&imgrefurl=http://www.dbgweb.com/current_alpha_systems.htm&h=952&w=792&sz=39&hl=en&start=1&tbnid=_rJF15RR0Ez8eM:&tbnh=148&tbnw=123&prev=/images%3Fq%3Dserver%26gbv%3D2%26svnum%3D10%26hl%3Den

http://images.google.com.au/imgres?imgurl=http://img.alibaba.com/photo/11046034/Prime_Array_NetServ_CD___DVD___NAS_Server.jpg&imgrefurl=http://www.alibaba.com/catalog/11046034/Prime_Array_NetServ_CD_DVD_NAS_Server.html&h=367&w=700&sz=77&hl=en&start=11&tbnid=gHIyr-plK4NwOM:&tbnh=73&tbnw=140&prev=/images%3Fq%3Dserver%26gbv%3D2%26svnum%3D10%26hl%3Den

http://images.google.com.au/imgres?imgurl=http://www.engadget.com/media/2006/05/dell-server.jpg&imgrefurl=http://www.engadget.com/2006/05/18/dell-goes-amd-in-servers-at-least/&h=250&w=183&sz=12&hl=en&start=7&tbnid=9uI1LtLWsFSiTM:&tbnh=111&tbnw=81&prev=/images%3Fq%3Ddell%2Bserver%26gbv%3D2%26svnum%3D10%26hl%3Den






4

Challenges

One example of why is it important to a Bank

Input Output Nomenclature

Bootstrapping

Interpolation

Perturbations

Bootstrap Perturbed Curve

Interpolation

Perturbed

Interpolated

Curve

+

CurveRates Data

+

Curve+

Curve

Rates Data

Rates Data

Curve

Pri

ce

Cash Future Swap

1 D

ay B

ill

1 M

on

th B

ill

2 M

on

th B

ill

3 M

on

th B

ill

Futu

re

6 m

th

Futu

re

12

mth

Futu

re

18

mth

Futu

re

24

mth

3 Y

ear

4 Y

ear

5 Y

ear

7 Y

ear

10

Ye

ar

15

Ye

ar

25

Ye

ar

For example: Value at Risk (a regulatory risk control)

Minimum 2 years of daily history

Over 5 Risk Factors

= ~ 500 (trading days) * 5 (risk factors) * entire trade population

= 2,500 valuations per trade each time to calculate


5

Benefits and Return On Investment

Calypso

Commercial Off the Shelf Implementation

The time to install the TIBCO DataSynapse GridServer components on the NAB Test Environment took approximately two hours.

The time taken to install and deploy the Calypso integration layer was approximately two hours

To the limits of the available hardware the scalability of the resulting grid-enabled Calypso deployment is almost completely linear for on-grid processing.

Results

In pre-grid testing, the Calypso Scheduling engine reached a capability limit. With grid processing this instability was removed.

Calypso task preparation and collection contains a bottleneck which skews end to end performance as measured from Calypso. For example, the results are written to a file by a single process.

Near linear scalability across the grid

PLUS:

• Expanded full portfolio recalculation, previously limited due to capacity constraints

Now: Live and expanding


6

Benefits and Return On Investment

BGS Provision Model:

Excel based Monte Carlo simulation:

The BGS Provision Model is delivered as a set of C++ dlls

The entry point method accepts as input a set of data, including Trades,

Schedules, Scenarios, Curves, Volatilities, etc.

The calculation is a 3-factor Monte Carlo simulation.

After each calculation the dlls return a vector of values

Spreadsheet

• performs the shifting operations,

• marshals the input data and calls the entry point method of the BGS Provision

Model dlls 7 times; once for each of the interest rate shifts.

• The dll invocation is effected directly from Visual Basic code

Desire to increase to a minimum of 512 shifts, each containing 50,000 paths.

Linearly, this would increase the elapsed time for the calculations in excess of 300

times,

Pre-grid, this would take a minimum run time on a single PC of over 50 hours!


Architecture


8

The Master Worker Pattern

http://www4.java.no/presentations/javazone/2006/slides/4623.pdf

http://www4.java.no/presentations/javazone/2006/slides/4623.pdf


9

Implementation

TIBCO GridServer

additional benefits:

Deployment of code

managed centrally

“Profile” matching:

Hardware, runtimes,

libraries, etc

Ease of deployment

Audit trail of usage (i.e.

chargeback capability)


10

A shift in paradigm

Non-grid

Monitor CPU

Capacity must be sized for maximum demand• If reach maximum demand risk catastrophic failover

Grid

Not possible to overwhelm workers• Only take work when available

• Can reliably work at 100% utilisation

• If worker fails, grid manager re-assigns task

• Focus;

sustained put that exceeds take


11

Best Practices


12

Candidates

Grid Server

“Model” implementations• Look for the “for” loops

• Be aware, it will expose bottlenecks and design flaws – this is a good thing!

Master Worker pattern implementations• Anything that can be broken into tasks and carried out in parallel

Breaking the back of Batch

Fabric Server

Server deployments• Extreme availability websites

• “Peaky” applications


13

Lessons Learned

Difficult to touch and feel

Baby steps


14

Wrap Up and Summary

Consider a multi-dimensioned business case:

Business Sponsors

• More comprehensive decision analytics; Revenue/Risk drivers

Technology Leadership

• Cost reduction; Less servers -> the no maintenance utopia

• Resilience

• Quality of Service

• Service charge models, pay for capability not server

Corporate Social Responsibility

http://intranet.global.thenational.com/News/HotTopics/2009/October/Pages/FullYearResults2009.aspx