Kanban for large scale off-shored maintenance @ mobile.de January 26 2010 - Munich Feedback to mandrezak@team.mobile.de
feedback for all things Kanban at OOP 2010: #oopkanban
small bugs, small and medium features Photo by dolorix - http://flic.kr/p/5mtavt
national vs. international: dedicated team Photo Nasa, visible earth
hard to manage, lots of WIP, hierarchical culture
cycle time = 28 days lead time = 52 days
you want half of these times? you have specialized teams? they are distributed?
Enter Kanban! (signal card)
Photo by john w - http://flic.kr/p/8jcfa
10 Photo Sune Gynthersen, bestbrains.dk
value stream mapping
Input Development QA Production Ongoing Done Ongoing Done
design of Kanban cards
tasks / stories are represented as cards
map cards to value stream on the wall
Input Deelopment QA Production Ongoing Done Ongoing Done
definition of Work-In-Progress-limits (WIP-Limits) & Pull
Input Development Acceptance 5 5 Ongoing Done Ongoing Done Produktion Pull Fixed WIP-Limits in relevant phases
definition of input and output-cadence
e.g. weekly prioritization e.g. bi-weekly releases Input Development Acceptance 5 5 Ongoing Done Ongoing Done Production Pull Fixed WIP limits for relevant phases
definition of done per phase and / or rules on pulling
Kanban cards visualize overall process state and clearly indicate bottlenecks in the process ---> basis for applying TOC
Theory of Constraints (light) - Eliah Goldratt: identify bottleneck max. stress on bottleneck eliminate (or minimize) bottleneck Eliah Goldratt - The Goal
definition of WIP limits reduction of batch size reduction of lead time (Little s law)
Little s Law: lead time = WIP / throughput easy to change: discipline / Flow hard to change: mid term change process
estimations ---> calculation of cycle time and lead time
@
Maintenance - own track
why does Kanban match our requirements?
Chasin! low visibility my project Let s try! Photo by Balakov - http://flic.kr/p/6knucl
Things are slow I like it! Photo by Dunechaser - http://flic.kr/p/jvebt
estimations not required
frequent prioritization! -> weekly production incidents any time
Output Queue flexibly coupled to release cycle and hotfixes
how did we implement it?
Photo by Dunechaser - http://flic.kr/p/agznc Semi-dictator style
I wanted desperately
first process draft
Input Analyse Development 2 2 4 5 QA Published Deployed Ongoing Done Ongoing Done Ongoing Done
feature branches! Photo by danorbit. - http://flic.kr/p/3lj2b9
training via Photo by pt - http://flic.kr/p/vddv
vacation :-( Photo by wili_hybrid - http://flic.kr/p/56vmt7
QA doesn t join the flow Photo by fsudaft - http://flic.kr/p/5diufn
not enough integration systems Photo by bobthemtnbiker - http://flic.kr/p/5jpnpm
QA w/ limited availability -> buffer Photo by rev - http://flic.kr/p/6hwer3
Entwicklung QA 5 5 Entwicklung ready f. QA QA 5 10 2 Ongoing Done Ongoing Done
flow! Photo by ViaMoi - http://flic.kr/p/6qlxqz
sucess
Cycle Time 28 24 30,0 22,5 15,0 august september 8 october 2 november 7,5 0 49
Lead Time (2-3 weekly releases) 52 60,0 45,0 august 28 september 22 october 30,0 15,0 0 50
but: once Chaos - now controlled
good mood in the team through flow experience
great communication tool
heavy discussion with uninvolved great buy-in with those involved
maintenance of whole platform from Jan 2010 w/ Kanban
1PO 1 Technical Lead 1 process guy 7 developers
Kanban adjustments required
introduction of classes of service for production incidents required / SLA
Thanks for your time and attention! contact: mandrezak@team.mobile.de twitter.com/markusandrezak slideshare.com/mandrezak
more on Kanban 5.3 - Thu 14:30-15:30 Stefan Roock & Bernd Schiffer Was ist dran an Kaban Feedback: #oopkanban 02/2010-26. Feb 2010 articles by Stefan Roock & Markus Andrezak