Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
An	
  Exploratory	
  Study	
  of	
  the	
  Evolu4on	
  of	
  
Communicated	
  Informa4on	
  
about	
  the	
  Execu4on	
  o...
2	
  
What	
  run-­‐4me	
  ac4ons	
  cause	
  the	
  
failure?	
  
Automated	
  profiling	
  &	
  instrumenta4on	
  
Detail	
   No	
  domain	
  knowledge	
  Large	
  scale	
  
3	
  
Communicated	
  informa4on	
  (CI)	
  
Execu4on	
  	
  
Logs	
  
System	
  
Alerts	
  
Code	
  
Comments	
  
/*	
  
…	
  
...
CI	
  forms	
  basis	
  of	
  Ecosystem	
  of	
  Log	
  Processing	
  
Apps	
  
Workload	
  recovery	
  Anomaly	
  
detec4...
How	
  to	
  keep	
  Log	
  Processing	
  Apps	
  in	
  sync	
  with	
  
CI?	
  
Release	
  1	
   Release	
  2	
   Release...
Our	
  Study	
  Dimensions	
  
What	
  types	
  of	
  
modifica*ons	
  
happen	
  to	
  CI?	
  
What	
  informa*on	
  is	
 ...
Case	
  Study	
  Setup	
  
Data	
  
Collec*on	
  
Log	
  
Abstrac*on	
  
System	
  
Deployment	
  
*me=1,	
  Trying	
  to	...
Our	
  Study	
  Dimensions	
  
What	
  types	
  of	
  
modifica*ons	
  
happen	
  to	
  CI?	
  
What	
  informa*on	
  is	
 ...
CI	
  keeps	
  on	
  growing	
  over	
  4me	
  
0	
  
20	
  
40	
  
60	
  
80	
  
100	
  
120	
  
140	
  
160	
  
180	
  
...
…even	
  when	
  system	
  size	
  decreases	
  
#	
  K	
  SLOC	
   #	
  Execu*on	
  log	
  events	
  	
  
0.19.0	
   293	...
CI	
  is	
  impacted	
  by	
  re-­‐engineering	
  
0.00%	
  
20.00%	
  
40.00%	
  
60.00%	
  
80.00%	
  
100.00%	
  
0.15....
How	
  does	
  CI	
  
evolve	
  over	
  
*me?	
  
13	
  
Growing	
  &	
  
changing	
  
	
  
Document	
  &	
  
track	
  
Wh...
Six	
  types	
  of	
  modifica4on	
  exist	
  
Rephrasing	
   Redundant	
  
informa4on	
  
	
  
Adding	
  
informa4on	
  
	...
Six	
  types	
  of	
  modifica4on	
  exist	
  
Rephrasing	
   Redundant	
  
informa4on	
  
	
  
Adding	
  
informa4on	
  
	...
Six	
  types	
  of	
  modifica4on	
  exist	
  
Rephrasing	
   Redundant	
  
informa4on	
  
	
  
Adding	
  
informa4on	
  
	...
Six	
  types	
  of	
  modifica4on	
  exist	
  
Rephrasing	
   Redundant	
  
informa4on	
  
	
  
Adding	
  
informa4on	
  
	...
Six	
  types	
  of	
  modifica4on	
  exist	
  
Rephrasing	
   Redundant	
  
informa4on	
  
	
  
Adding	
  
informa4on	
  
	...
Six	
  types	
  of	
  modifica4on	
  exist	
  
Rephrasing	
   Redundant	
  
informa4on	
  
	
  
Adding	
  
informa4on	
  
	...
Six	
  types	
  of	
  modifica4on	
  exist	
  
Rephrasing	
   Redundant	
  
informa4on	
  
	
  
Adding	
  
informa4on	
  
	...
Most	
  modifica4ons	
  can	
  be	
  avoided	
  
9.86%	
  
61.97%	
  
14.08%	
  
7.04%	
   7.04%	
   2.82%	
  
0.00%	
  
10...
How	
  does	
  CI	
  
evolve	
  over	
  
*me?	
  
22	
  
Growing	
  &	
  
changing	
  
	
  
Document	
  &	
  
track	
  
Wh...
Short-­‐lived	
  CI	
  contains	
  implementa4on	
  details	
  
Hadoop	
  saves	
  output	
  to	
  a	
  machine.	
  
Hadoo...
How	
  does	
  CI	
  
evolve	
  over	
  
*me?	
  
24	
  
Growing	
  &	
  
changing	
  	
  
	
  
Document	
  &	
  
track	
 ...
25	
  
Upcoming SlideShare
Loading in …5
×
Upcoming SlideShare
Msr2016 tarek
Next
Download to read offline and view in fullscreen.

0

Share

Download to read offline

WCRE2011

Download to read offline

WCRE2011

Related Books

Free with a 30 day trial from Scribd

See all

Related Audiobooks

Free with a 30 day trial from Scribd

See all
  • Be the first to like this

WCRE2011

  1. 1. An  Exploratory  Study  of  the  Evolu4on  of   Communicated  Informa4on   about  the  Execu4on  of  Large  So=ware  Systems     Weiyi Shang Zhen Ming Jiang Bram Adams Ahmed E. Hassan Michael W. Godfrey University of WaterlooQueen’s University   Mohamed Nasser Parminder Flora Research In Motion (RIM)
  2. 2. 2   What  run-­‐4me  ac4ons  cause  the   failure?  
  3. 3. Automated  profiling  &  instrumenta4on   Detail   No  domain  knowledge  Large  scale   3  
  4. 4. Communicated  informa4on  (CI)   Execu4on     Logs   System   Alerts   Code   Comments   /*   …   */   Sta*c  Dynamic   4   Field  experience  Developer  experience  
  5. 5. CI  forms  basis  of  Ecosystem  of  Log  Processing   Apps   Workload  recovery  Anomaly   detec4on   Capacity   planning   System   monitoring   Performance   analysis   5   Failure   diagnosis  
  6. 6. How  to  keep  Log  Processing  Apps  in  sync  with   CI?   Release  1   Release  2   Release  3   6  
  7. 7. Our  Study  Dimensions   What  types  of   modifica*ons   happen  to  CI?   What  informa*on  is   conveyed  by  the    short-­‐lived  CI?   Quan4ty   Type   Content   How  does  CI   evolve  over   *me?   7  
  8. 8. Case  Study  Setup   Data   Collec*on   Log   Abstrac*on   System   Deployment   *me=1,  Trying  to  launch,  TaskID=01A     *me=$t,  Trying  to  launch,  TaskID=$id     Enterprise  Applica*on  (EA)   8   Log   Events  
  9. 9. Our  Study  Dimensions   What  types  of   modifica*ons   happen  to  CI?   What  informa*on  is   conveyed  by  the    short-­‐lived  CI?   Quan4ty   Type   Content   How  does  CI   evolve  over   *me?   9  
  10. 10. CI  keeps  on  growing  over  4me   0   20   40   60   80   100   120   140   160   180   0.14.0   0.15.0   0.16.0   0.17.0   0.18.0   0.19.0   0.20.0   0.20.1   0.20.2   0.21.0   releases   #   execu*on     events   10  
  11. 11. …even  when  system  size  decreases   #  K  SLOC   #  Execu*on  log  events     0.19.0   293     113   0.20.0   250   121   11  
  12. 12. CI  is  impacted  by  re-­‐engineering   0.00%   20.00%   40.00%   60.00%   80.00%   100.00%   0.15.0   0.16.0   0.17.0   0.18.0   0.19.0   0.20.0   0.20.1   0.20.2   0.21.0   Unchanged  CI   Large  amounts  of  implementa4on  changes   12  
  13. 13. How  does  CI   evolve  over   *me?   13   Growing  &   changing     Document  &   track   What  types  of   modifica*ons   happen  to  CI?   What  informa*on  is   conveyed  by  the    short-­‐lived  CI?   Quan4ty   Type   Content  
  14. 14. Six  types  of  modifica4on  exist   Rephrasing   Redundant   informa4on     Adding   informa4on     Dele4ng   informa4on     Diverging     Merging   14  
  15. 15. Six  types  of  modifica4on  exist   Rephrasing   Redundant   informa4on     Adding   informa4on     Dele4ng   informa4on     Diverging     Merging   Hadoop  mapred  Reduce  task  fetch  n  bytes     Hadoop  MapReduce  task  Reduce  fetch  n  bytes     15  
  16. 16. Six  types  of  modifica4on  exist   Rephrasing   Redundant   informa4on     Adding   informa4on     Dele4ng   informa4on     Diverging     Merging   ShuffleRamManager  memory  limit  n  MaxSingleShuffleLimit  m   ShuffleRamManager  memory  limit  n  MaxSingleShuffleLimit  m   mergeThreshold  Q   16  
  17. 17. Six  types  of  modifica4on  exist   Rephrasing   Redundant   informa4on     Adding   informa4on     Dele4ng   informa4on     Diverging     Merging   Adding  task  to  tasktracker     Adding  Map  Task  to   tasktracker   Adding  Reduce  Task  to   tasktracker   17  
  18. 18. Six  types  of  modifica4on  exist   Rephrasing   Redundant   informa4on     Adding   informa4on     Dele4ng   informa4on     Diverging     Merging   Avoidable   18  
  19. 19. Six  types  of  modifica4on  exist   Rephrasing   Redundant   informa4on     Adding   informa4on     Dele4ng   informa4on     Diverging     Merging   Recoverable   19  
  20. 20. Six  types  of  modifica4on  exist   Rephrasing   Redundant   informa4on     Adding   informa4on     Dele4ng   informa4on     Diverging     Merging   Unavoidable   20  
  21. 21. Most  modifica4ons  can  be  avoided   9.86%   61.97%   14.08%   7.04%   7.04%   2.82%   0.00%   10.00%   20.00%   30.00%   40.00%   50.00%   60.00%   70.00%   80.00%   90.00%   100.00%   redundant   info   rephrasing   adding  info   dele4ng   info   diverging   merging   avoidable   recoverable   unavoidable   21  
  22. 22. How  does  CI   evolve  over   *me?   22   Growing  &   changing     Document  &   track   What  types  of   modifica*ons   happen  to  CI?   What  informa*on  is   conveyed  by  the    short-­‐lived  CI?   Quan4ty   Type   Content   6  types       Are  mostly   avoidable  
  23. 23. Short-­‐lived  CI  contains  implementa4on  details   Hadoop  saves  output  to  a  machine.   Hadoop  assigns  a  reduce  task  to  a  machine.   Map  task  updates  its  progress.   Hadoop  reads  from  a  local  file.   Hadoop  Abempt  saves  its  output  and  reports  to   the  task  tracker.   23   Node  name   Local  path   Using  ipc   Output  file  name  
  24. 24. How  does  CI   evolve  over   *me?   24   Growing  &   changing       Document  &   track   What  types  of   modifica*ons   happen  to  CI?   What  informa*on  is   conveyed  by  the    short-­‐lived  CI?   Quan4ty   Type   Content   6  types       Are  mostly   avoidable   Implementa*on-­‐ level  details     Fragile   Maintenance   effort  
  25. 25. 25  

WCRE2011

Views

Total views

24,313

On Slideshare

0

From embeds

0

Number of embeds

23,386

Actions

Downloads

1

Shares

0

Comments

0

Likes

0

×