alex furneaux december 2011 - dan-lewis … · capgemini provides the national reservations service...
TRANSCRIPT
Alex Furneaux
December 2011
BITUG
Transport
Contents
Who are ATOC?
Why change/upgrade?
Choices, choices and yet more choices
Price and that old sizing problem
Old for new applications
Results and benefits
Bottom Line : Why did we stay with Non-Stop?
Alex Furneaux, Dec 2011
2 © 2011 Capgemini. All rights reserved.
Transport
Who are ATOC? Formed in 1993 ‘As the voice of the passenger railway, ATOC represents train companies to the government, regulatory bodies, the media and other Opinion formers on transport policy issues’ Its work is administered through four companies: ATOC Limited, through which it provides most of its services, employs the staff and acts as agent for many of the schemes; Rail Settlement Plan Limited (RSP), which manages the revenue; Rail Staff Travel Limited, managing staff travel, and National Rail Enquiries, the public information provider. Capgemini provides the National Reservations Service to RSP and its members, who included on-line retailers and the Train Operating Companies (TOC’s)
Alex Furneaux, Dec 2011
3 © 2011 Capgemini. All rights reserved.
Transport
Rail Settlement Plan (RSP) provides central retail support services to the UK Train Operating Companies (TOCs). Services include;
Collects retail sales data from 10,000 points-of-sale
Carries out the correct allocation of ticket revenue to train operators
Settles that revenue to the operators on a daily basis
Sets standards and accredits all industry points of sale
Maintains the central industry fares database and provides tools for train operators to set fares
Distributes fares, timetable, station and other industry data to points of sale/information systems
Provides the NRS enabling retailers to book reservable seat on all trains
Provides the capability to pick up pre-ordered tickets at station ticket machines (TOD)
Provides the rail industry with ticket stock
Rail Settlement Plan - Business
Alex Furneaux, Dec 2011
4 © 2011 Capgemini. All rights reserved.
Transport
RSP procures these services on behalf of the TOCs and manages their operation via third party suppliers. Capgemini is one of these suppliers, along with ATOS Origin, Fujitsu and Thales. RSP itself is small company formed of approx 27 staff, based in London. Annual turDecer is approximately £30m. They also have responsible for the annual allocation of £4.8 billion of rail revenue monies to the TOC’s.
Rail Settlement Plan (Contd.)
Alex Furneaux, Dec 2011
5 © 2011 Capgemini. All rights reserved.
Transport
NRS (National Reservation System)
Key benefits of NRS;
• Improved overall functionality
• Faster and easier to use
• Increased flexibility to review availability
and booking seats
• Improved functionality for disabled
passengers to reserve seats and reserve
assistance at stations
• Improved management information
regarding inventory & ability to apply new
revenue management techniques to
maximise yields.
Initiated in Jan 1999
Alex Furneaux, Dec 2011
6 © 2011 Capgemini. All rights reserved.
Transport
NRS Volumes (~ 2008)
Servicing a “business” with
• 2,500 stations
• 4,000 trains and 20,000 services
• circa 2.75 million passengers every day
The NRS Service
• 193,161 services stored
• 50.41 million accommodation units
• 197 transactions per second (tps)
• 80.1million reservations held (2007/8)
The original system went live in Dec
2004.
Alex Furneaux, Dec 2011
7 © 2011 Capgemini. All rights reserved.
Transport
Original RMS System
Alex Furneaux, Dec 2011
8 © 2011 Capgemini. All rights reserved.
Transport
NRS Volume growth
Alex Furneaux, Dec 2011
9 © 2011 Capgemini. All rights reserved.
Transport
The Problem
The original system was implemented in Dec 2004 and “running out of
steam”
• Based on HP Tandem S86000 (fully loaded)
• Designed to process ~ 200 tps
• Caching servers added to provide additional performance (up to ~ 500 tps)
• Still cannot meet growing demand of the service
• The batch window was increasingly unable to meet demand
• Additional services and TOCs being added simply fuelled the performance
issues
More performance with faster throughput was required
Hardware was running out of extended support and was fully depreciated on
RSP’ books.
Alex Furneaux, Dec 2011
10 © 2011 Capgemini. All rights reserved.
Transport
However…
With business change (service alterations), we came to a “tipping
point” – the system could not sustain the projected growth
The challenge was to find;
• Cost effective solution – review all cost smoothing options over life of the deal
• Minimal change to the overall infrastructure
• Scalable – something that provided an upgrade path with the capability to
reach 5,000 tps if required
• Continuance of service (24*7) being a critical factor
• Minimal application change
and;
18 + TOCs plus Retailers had to agree to the funding of any
change.
Alex Furneaux, Dec 2011
11 © 2011 Capgemini. All rights reserved.
Transport
Choices, choices
Stay with HP Tandem – but, we know non-stop servers and software are
expensive (Outrageous – blame HP )
Move away from the HP platform
• Move to another tier-1 production system – Oracle’s ExaData for example
• Use a LINUX/low cost operating system environment
• Develop a series of services to access data from a production database
back-end system
Develop an “As-a-Service” offering to avoid any large CAPEX
CONCLUSION
Following an in depth review of the alternatives with RSP it was agreed we
would stay with HP NSK and build on the success we had had to date
Reviewed various options within the NSK range – 2000, 16000 & 50000
• As these were new range of processors there were very few performance
stats available, which became an issue later on.
Alex Furneaux, Dec 2011
12 © 2011 Capgemini. All rights reserved.
Transport
Architectural Considerations
The current solution was fundamentally sound
Any major change/disruption would have had a big impact on
TOC’s and online retailers if things went wrong – loss of revenue
and credibility
Therefore;
We required a long-term solution without incurring major change
to the infrastructure or application
Had to be at the best price point, both on a CAPEX and a longer
term TCO basis and at minimum risk
Flexible, so we could upgrade when required – don’t buy all at
once, only when required
Major coding changes were to be avoided at all cost.
Alex Furneaux, Dec 2011
13 © 2011 Capgemini. All rights reserved.
Transport
Performance, Price & TPS
One of our major considerations was getting the initial sizing of
the NSK correct as any mistake would have been very costly
• Moving from 1 enclosure to 2 results in a large price point hike
• Blades are purchased in pairs so had to ensure they were required
• Big hike in software & maintenance costs
Identifying key change price points as we scaled up to 5,000 tps
If possible, retain as much of the old infrastructure as possible
“Real life testing” was incredibly difficult
• Couldn’t run it on a “similar” test machine (not around)
• Would be expensive to build a “test” rig
• Real vs calculated quandary
One of the longest timelines in the project.
Alex Furneaux, Dec 2011
14 © 2011 Capgemini. All rights reserved.
Transport
Performance testing/estimating
Alex Furneaux, Dec 2011
15 © 2011 Capgemini. All rights reserved.
Transport
New Outline Architecture – based on analysis
New
Alex Furneaux, Dec 2011
16 © 2011 Capgemini. All rights reserved.
Transport
Indications from the Analysis
The mix of OLTP & batch issues caused major issues with the overall loading of the system
5,000 tps was possible with the largest NB50000 (2 racks) • Scalable, flexible, no major code re-writing
• Unfortunately, it was too expensive overall
Tipping point to step-change was around 3,000 tps where a move to a second rack was estimated to be required
Change the batch process on the NSK
Architecturally, a reduction in overall NSK load was required.
This was reviewed and agreed with RSP.
Alex Furneaux, Dec 2011
17 © 2011 Capgemini. All rights reserved.
Transport
Performance Results – reality intervenes
Used an NB2004 to “firm up” performance estimates • later used as the test machine in the new configuration
With the speed of the new processors greatly improving throughput, we removed the caching servers from the overall solution
A re-compile of the application, along with some minor code changes, this provided excellent performance results
Our initial estimates proved to be conservative • TPS figures were roughly double our original estimates
Avoided the need for a 2nd rack for the higher tps levels.
Alex Furneaux, Dec 2011
18 © 2011 Capgemini. All rights reserved.
Transport
Final Outline Architecture
Alex Furneaux, Dec 2011
19 © 2011 Capgemini. All rights reserved.
Transport
Summary – So did we stick with Non-Stop?
Performance
Reliability (1 outage in some 6 years – mainframe class levels)
Long term TCO is good
Reduced support requirements
Cheaper than a port to an alternative supplier platform
Flexible solution, enabling RSP to develop new services and
increase performance/capacity if/when required
Adapts the solution to web services and the online marketplace
• Trainline.com, RedspottedHanky.com, etc.
Alex Furneaux, Dec 2011
20 © 2011 Capgemini. All rights reserved.
www.capgemini.com
The information contained in this presentation is proprietary. ©2011 Capgemini. All rights reserved