PADRES

This a presentation of PADRES (Publish/subscribe Applied to Distributed REsource Scheduling), a publish/subscribe middleware directed towards workflow management and business process execution.
The presentation was made for "Middleware Technologies" couse, for my Master in Information Security course, FCUL.

Transcript

1. PADRES   Publish/subscribe  Applied  to  Distributed  REsource  Scheduling     G.  Li,  A.  Cheung,  Sh.  Hou,  S.  Hu,  V.  Muthusamy,     R.  Sherafat,  A.  Wun,  H.-­‐A.  Jacobsen,  S.  Manovski       Presenta(on  by  Tiago  Posse  
2. What  is  PADRES   •  Distributed  P/S   •  Content-­‐based   •  Directed  towards  workflow  management  and  business   process  execu(on    
3. Why  PADRES   •  Numerous  publish/subscribe  technologies   •  LiBle  aBen(on  to  workflow  management  and  business   process  execu(on   •  Requirements   •  Access  to  past  events   •  Content-­‐based  rou(ng   •  Express  the  complicated  dependencies  in  a  workflow  using  a  subscrip(on   language  
4. PADRES   •  Set  of  interconnected  brokers   •  ORT   •  Rete  engine   •  Main  features:   •  Access  to  historic  data   •  Composite  subscrip(ons  
5. Messaging   •  Adver(se-­‐publish-­‐subscribe   •  [aBribute,  operator,  value]  tuples   •  Classes   •  Class  granularity  
6. Advertisements   •  Adver(se  -­‐>  publish   •  Subscrip(on  Rou(ng  Table   •  Flooding   •  Merging       B2   B1   B3   B4   P   1)  Adv  (x)   1)  Adv  (x)   1)  Adv  (x)   1)  Adv  (x)   SRT  of  B1   Adver&sement   Last  hop   Adv  (x)   P   SRT  of  B4   Adver&sement   Last  hop   Adv  (x)   B2  
7. Subscriptions   •  Rou(ng  through  SRT   •  Publica(on  Rou(ng  Table   •  Merging   B2   B1   B3   B4   P   1)  Adv  (x)   1)  Adv  (x)   1)  Adv  (x)   1)  Adv  (x)   S   2)  Sub  (x)   2)  Sub  (x)   2)  Sub  (x)   PRT  of  B3   Subscrip&on   Last  hop   Sub  (x)   S   PRT  of  B4   Subscrip&on   Last  hop  
8. Publications   •  Rou(ng  through  PRT’s   B2   B1   B3   B4   P   S   2)  Sub  (x)   2)  Sub  (x)   2)  Sub  (x)   3)  Pub  (x)   3)  Pub  (x)   3)  Pub  (x)   3)  Pub  (x)  
9. Access  to  historic  data   •  Historic  databases   •  Re-­‐publish  old  publica(ons   •  Transparent  to  par(cipants     •  Addi(on  of  (me  predicate   •  Historic  subscrip(ons   •  Hybrid  subscrip(ons  
10. Composite  subscriptions   •  Several  atomic  subscrip(ons   •  Logical  and/or  temporal  connec(ons   •  Ordered  matching   •  Subscrip(on  spliZng   •  Distributed  composite  event  matching  
11. Rete  engine   •  One  input  queue   •  Several  output  queues   •  Rete  tree   •  Par(al  matching  state   •  4.5  ms  to  test  200.000  subscrip(ons   •  80x  faster  than  linear  algorithm   •  20x  faster  than  coun(ng  algorithm  
12. Rete  network   •  Check  event  again  predicates   •  Put  into  queue  when  matched   •  Par(ally  matched  state   Root   Class!=stock   Class=stock   …   Price<20   Price>50   Queue  2   Queue  1   join   Queue  3   Sub  1   Sub  2   CS   Publica(on  1   <class,=,stock>   <price,=,60>   Matched,  so  goes  down   Check  for  matches   Put  into  queue  to  be   delivered  to  last  hop   Par(ally     matching   state   Publica(on  2   <class,=,stock>   <price,=,10>   CS  is  matched   Put  into  queue  
13. PADRES   •  Advantages:   •  Access  to  old  publica(ons   •  Powerful  subscrip(ons  with  distributed  event  matching     •  Suitable  for  large  scale  p/s:  Rete  engine   •  Problems:   •  Classes   •  Placement  and  content  of  historic  databases   •  Message  cycles  
14. Message  cycles   •  Pub(x)  overlaps  with  both  Sub(x)  and  Sub(y)   B2   B3   B1   Sub(y)   Sub(x)   Sub(x)   Sub(y)   Sub(x)   Pub  (x)   Subscrip&on   Last  hop   Sub  (x)   B3   Sub(y)   B6   Subscrip&on   Last  hop   Sub  (x)   B1   Subscrip&on   Last  hop   Sub  (x)   B5   Sub(y)   B2   SRT  of  B2   SRT  of  B3   SRT  of  B1   CYCLE!  
15. Questions?