ProductiZing Research

1
ProductiZing Research
John L. Miller, Software Architect Microsoft
Research Cambridge 11-July 2007
2
Agenda

• What is Productization?
• Why is it difficult?
• Can I do it myself?
• Advice from the field

3
What is Productizing

• Migrating research results
• Affecting an existing product
• Supplying algorithms and methods
• Influencing product features
• Writing shipping code
• Seeding a new product

4
BM25F An algorithm

• Algorithm for web search ranking
• Devised at MSRC
• Results transferred to MS search teams
• Product team wrote their own version based upon
  the research algorithm

5
Sideshow a feature

• Gadget Bar for Windows
• Prototyped shared at Microsoft years ago
• Now part of Vistathe Sidebar
• Product team wrote their own code
• strikingly similar appearance and features

6
TrueSkillTM Shipping Code

• Player Ranking and Matchmaking
• Researched in MSRC
• Primary code developed by researchers
• Incorporated with some modifications
• Basis of matchmaking for most Xbox-360 games

7
Pastry external licensing

• Pastry / Splitstream / Scribe / Squirrel
• P2P communication framework, applications
• Shopped inside of Microsoft, but never
  successfully transferred
• Licensed to Skinkers for spin-out
• LiveStation in Beta, recently in the press

8
The visible minority

• MSRC has dozens of successful tech transfers
• Many more never started, or fizzled out
• No appropriate target
• Best fit already too busy
• Change in direction / management
• Lack of resources

9
Why is Productization difficult?

• Orders of magnitude more work than research
• It cannot fail vs. it can succeed
• Requires influencing people you dont have any
  control over
• Internal work is a priority
• Existing teams already have lots to do
• Understanding new technology takes work

10
Productizing PNRP timeline
Jan 01
Jan 02
Jan 03
1-Algorithm
2-Prototyped
3-Designed
4-Implemented
5-Tested
6-Released

• 2000-10 PNRP proposed as algorithm by Huitema
• 2001-03 Prototype completed
• 2 person-months effort
• Fully functional IPv4 prototype
• Broader than final product, but not as deep
• 2002-04 Initial design document first draft
  completed
• 2-3 person-years effort
• 40 page protocol spec, 150 page design spec
• 2002-08 Code-complete
• 1 person-year effort for dev, 1/2 person year
  for test
• 2003-04 Test-complete, ship bugs fixed
• 1.5 person-years effort for dev, 2 person -years
  effort for test
• 2003-06 Released to Web (RTW)

11
Product Quality

• Strict coding guidelines
• Modularity, extensibility, scalability
• Diagnostics
• Robustness
• Prototype
• Design for majority case
• Failure exit with apologies
• Access violation fix, re-run
• Product code
• Design for ALL cases
• Failure work around the problem
• Access violation angry customers gt
  bulletproofing

12
Example Prototype
void DrawBox(HDC dc, int x, int y) bool
bGot false HPEN p, op if (dc
NULL) dc GetDC(NULL) bGot true
p CreatePen(PS_SOLID, 10, RGB(0, 255,0))
op SelectObject(dc, p) Rectangle(dc, x,
y, 200, 200) SelectObject(dc, op)
DeleteObject(p) if (bGot) ReleaseDC(dc)
13
Example Product Quality
// Pen of 10 pixels for drawing box const int
MY_PEN_WIDTH 10 // Green pen for our box
color const COLORREF GREEN_PEN_RGB RGB(0,
255,0) // All boxes we draw are 200 x 200 const
int MY_BOX_WIDTH 200 const int MY_BOX_HEIGHT
200 HRESULT DrawBox( IN HDC
hContext, IN int iXOffset, IN int
iYOffset ) BOOL fFreeDC FALSE,
fRet FALSE int iRet 0 HRESULT hrRet
S_OK HPEN hpNew NULL, hpOld NULL
// // Get the default screen DC if we don't
already have one. // if (NULL
hContext) hContext GetDC(NULL)
if (NULL hContext)
hrRet HRESULT_FROM_WIN32(GetLastError(
)) TraceMessage(MYAPP_ERROR,
L"s(d) - Unable to get screen DC 0x08x",
__TFUNCTION__, __LINE__, hrRet) goto
Cleanup fFreeDC
TRUE
// // Create a green pen, and select it
to draw our rectangle // hpNew
CreatePen(PS_SOLID, MY_PEN_WIDTH,
GREEN_BRUSH_RGB) if (NULL hpNew)
hrRet E_FAIL TraceMessage(MYAPP_E
RROR, L"s(d) - Unable to create pen 0x08x",
__TFUNCTION__, __LINE__, hrRet) goto
Cleanup hpOld SelectObject(hContext,
hpNew) if (NULL hpOld)
hrRet E_FAIL TraceMessage(MYAPP_ERROR,
L"s(d) - Unable to select our new pen 0x08x",
__TFUNCTION__, __LINE__, hrRet) goto
Cleanup // Draw the actual rectangle
// fRet Rectangle(hContext, iXOffset,
iYOffset, MY_BOX_WIDTH, MY_BOX_HEIGHT) if
(FALSE fRet) hrRet
HRESULT_FROM_WIN32(GetLastError())
TraceMessage(MYAPP_ERROR, L"s(d) - Unable to
Draw the rectangle 0x08x", __TFUNCTION__,
__LINE__, hrRet) goto Cleanup

Cleanup // Clean up the new brush, and
select the old brush if there was one //
if (NULL ! hpNew) if (NULL !
hpOld) HPEN hpTemp
SelectObject(hpOld) if (NULL !
hpTemp) hrRet
HRESULT_FROM_WIN32(GetLastError())
TraceMessage(MYAPP_ERROR, L"s(d) - Unable to
reset Pen to original 0x08x", __TFUNCTION__,
__LINE__, hrRet) else
hpOld NULL

DeleteObject(hpNew) hpNew NULL
// If we allocated a DC, free it. //
if (fFreeDC) iRet
ReleaseDC(hContext) if (0 iRet)
hrRet E_FAIL
TraceMessage(MYAPP_ERROR, L"s(d) - Unable to
release a default DC! Will be fatal soon
0x08x", __TFUNCTION__,
__LINE__, hrRet) else
hContext NULL
return(hrRet)
14
Other Productization work

• Funding, business model, management
• Testing, release management, product support
• Marketing, technical evangelism

15
Can I productize it myself?

• If youre a renaissance person, sure!
• A better question should you?
• Engineering not as easy as it looks
• LOTS of work
• Opportunity cost is high
• Forza 2 people x 2.5 years
• .NET Generics 1 person x 3 years

16
How does Microsoft Cope?

• Researchers focus on research
• RSDEs sometimes help with productization
• MSR Program Managers
• Introductions, relationship help
• Product / Research Events
• TechFest research showcase
• Mindswaps
• Other avenues for transfer
• External licensing
• Spinout

17
How to Succeed

• Identify your transfer goal
• Choose your technology carefully
• Pick the right target team
• Communicate
• Persistent and flexible
• Keep perspective

18
Identify your transfer goal

• Is the priority transferring your technology, or
  influencing a product?
• What direction will you accept?
• just in time research for team?
• Transfer your original research
• What is success?
• Team evaluates technology?
• Team adjusts course based on your work?
• Your research in the hands of users

19
Choose your technology

• Not all research is appropriate for
  productization
• How risky is it?
• How practical is it?
• How much work is it to adopt?
• What is the cost-benefit analysis?

20
Picking the right partner

• Does your work align with their goals?
• Do they have a good track record?
• Do they have a future?
• Is there mutual respect?
• Are they at the right stage in the product cycle?
• What is their commitment level?

21
Communicate

• Build / maintain trust and respect
• Identify potential issues before they become
  problems
• Identify key contributors, owners, and establish
  rapport
• Technical evangelism is an ongoing requirement
• Be flexible, adaptable, and patient

22
Keep perspective

• Its one project of many you will work on
• Its one project of many THEY will work on
• What are the chances its the most important work
  youll ever do?
• People outlive projects

23
Summary

• Productization is tough but worthwhile
• Set your own goals, understand other peoples
  goals
• Communicate, be persistent, and patient