In this paper we present a wide area distributed
system using a content-based publish/subscribe communication middleware
which can deterministically detect and report failures with respect to
timely message delivery and message omission. Our approach does not require
external clock synchronization nor does it impose any constraints on the
publish/subscribe middleware. We show that our system performs better and is
safer than when using NTP for external clock synchronization. We provide a
proof of concept implementation and present results of experiments carried
out in the PlanetLab environment.
Driving Behavioral Change for Information Management through Data-Driven Gree...
Fail-Aware Publish/Subscribe
1. Motivation Overview The Upper Bound Evaluation Summary 1 of 27 slides
Fail-Aware Publish/Subscribe
NCA 2007
Zbigniew Jerzak, Christof Fetzer, Robert Fach
Dresden University of Technology
October 16, 2008
{Zbigniew.Jerzak, Christof.Fetzer, Robert.Fach}@inf.tu-dresden.de
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
2. Motivation Overview The Upper Bound Evaluation Summary 2 of 27 slides
State-of-the-art
Problems:
we have conversation
we need dissemination
Dissemination:
you are interested in data (not source)
Hence loose-coupling:
anonymous
asynchronous
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
3. Motivation Overview The Upper Bound Evaluation Summary 2 of 27 slides
State-of-the-art
Problems:
we have conversation
we need dissemination
Dissemination:
you are interested in data (not source)
Hence loose-coupling:
anonymous
asynchronous
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
4. Motivation Overview The Upper Bound Evaluation Summary 2 of 27 slides
State-of-the-art
Problems:
we have conversation
we need dissemination
Dissemination:
you are interested in data (not source)
Hence loose-coupling:
anonymous
asynchronous
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
5. Motivation Overview The Upper Bound Evaluation Summary 3 of 27 slides
Content-Based Systems
Peer-to-Peer, Publish/Subscribe, CBN...
Content-/Name-driven
Loosely-coupled and distributed
Failures are imminent [MPHD06]
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
6. Motivation Overview The Upper Bound Evaluation Summary 3 of 27 slides
Content-Based Systems
Peer-to-Peer, Publish/Subscribe, CBN...
Content-/Name-driven
Loosely-coupled and distributed
Failures are imminent [MPHD06]
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
7. Motivation Overview The Upper Bound Evaluation Summary 4 of 27 slides
Fail-Awareness
...allows an application to [...] detect when it cannot guarantee all
its safety properties and in particular, to detect when to switch to
a fail-safe mode [FC03]
Did we receive all messages?
Was the received message timely?
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
8. Motivation Overview The Upper Bound Evaluation Summary 4 of 27 slides
Fail-Awareness
...allows an application to [...] detect when it cannot guarantee all
its safety properties and in particular, to detect when to switch to
a fail-safe mode [FC03]
Did we receive all messages?
Was the received message timely?
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
9. Motivation Overview The Upper Bound Evaluation Summary 5 of 27 slides
Publish/Subscribe Systems [CRW01]
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
10. Motivation Overview The Upper Bound Evaluation Summary 6 of 27 slides
Time in Publish/Subscribe
no clock synchronisation
no violation of loose-coupling
many-to-many communication
How long did it take for a message to be delivered from p to q?
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
11. Motivation Overview The Upper Bound Evaluation Summary 6 of 27 slides
Time in Publish/Subscribe
no clock synchronisation
no violation of loose-coupling
many-to-many communication
How long did it take for a message to be delivered from p to q?
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
12. Motivation Overview The Upper Bound Evaluation Summary 7 of 27 slides
Time in Publish/Subscribe
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
13. Motivation Overview The Upper Bound Evaluation Summary 8 of 27 slides
Idea
do not calculate the transmission time
calculate upper bound on transmission time
Message transmission time is no longer than the upper bound
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
14. Motivation Overview The Upper Bound Evaluation Summary 8 of 27 slides
Idea
do not calculate the transmission time
calculate upper bound on transmission time
Message transmission time is no longer than the upper bound
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
15. Motivation Overview The Upper Bound Evaluation Summary 9 of 27 slides
Upper Bound – Point to Point
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
16. Motivation Overview The Upper Bound Evaluation Summary 9 of 27 slides
Upper Bound – Point to Point
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
17. Motivation Overview The Upper Bound Evaluation Summary 9 of 27 slides
Upper Bound – Point to Point
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
18. Motivation Overview The Upper Bound Evaluation Summary 9 of 27 slides
Upper Bound – Point to Point
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
19. Motivation Overview The Upper Bound Evaluation Summary 9 of 27 slides
Upper Bound – Point to Point
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
20. Motivation Overview The Upper Bound Evaluation Summary 9 of 27 slides
Upper Bound – Point to Point
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
21. Motivation Overview The Upper Bound Evaluation Summary 9 of 27 slides
Upper Bound – Point to Point
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
22. Motivation Overview The Upper Bound Evaluation Summary 10 of 27 slides
Upper Bound – Publish/Subscribe
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
23. Motivation Overview The Upper Bound Evaluation Summary 10 of 27 slides
Upper Bound – Publish/Subscribe
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
24. Motivation Overview The Upper Bound Evaluation Summary 10 of 27 slides
Upper Bound – Publish/Subscribe
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
25. Motivation Overview The Upper Bound Evaluation Summary 10 of 27 slides
Upper Bound – Publish/Subscribe
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
26. Motivation Overview The Upper Bound Evaluation Summary 10 of 27 slides
Upper Bound – Publish/Subscribe
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
27. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
28. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
29. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
30. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
31. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
32. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
33. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
34. Motivation Overview The Upper Bound Evaluation Summary 11 of 27 slides
Upper Bound – Optimizations
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
35. Motivation Overview The Upper Bound Evaluation Summary 12 of 27 slides
Upper Bound vs NTP
NTP
no guarantees whatsoever
2*max error gives estimated upper bound
Upper Bound
precise
more accurate
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
36. Motivation Overview The Upper Bound Evaluation Summary 13 of 27 slides
Experiment Setup
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
37. Motivation Overview The Upper Bound Evaluation Summary 14 of 27 slides
TSC as Time Source
60
planetlab-1a.ics.uci.edu
planetlab01.mpi-sws.mpg.de
planetlab-2.cse.ohio-state.edu
50 planetlab2.exp-math.uni-essen.de
40
drift rate[ppm]
30
20
10
0
01:04 01:08 01:12 01:16 01:20 02:00 02:04 02:08 02:12 02:16
days:hours
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
38. Motivation Overview The Upper Bound Evaluation Summary 15 of 27 slides
TSC as Time Source – Single Host
6
planetlab-2.cse.ohio-state.edu
error
5.5
5
4.5
drift rate[ppm]
4
3.5
3
2.5
2
1.5
01:00 01:12 02:00 02:12 03:00 03:12 04:00 04:12 05:00 05:12
days:hours
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
39. Motivation Overview The Upper Bound Evaluation Summary 16 of 27 slides
Publication Transmission Time (national)
2.0e+05
F-A P/S upper bound
realtime
1.8e+05 error
omission error
1.6e+05
1.4e+05
omission error
time [us]
1.2e+05
1.0e+05
8.0e+04
6.0e+04
4.0e+04
yes
2.0e+04 no
0 500 1000 1500 2000 2500 3000
received publication #
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
40. Motivation Overview The Upper Bound Evaluation Summary 17 of 27 slides
Publication Transmission Time (global)
108
F-A P/S upper bound
realtime
error
omission error
107
omission error
time [us]
106
yes
105 no
0 500 1000 1500 2000
received publication #
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
41. Motivation Overview The Upper Bound Evaluation Summary 18 of 27 slides
Publication Transmission Time – Bootup (national)
2.0e+05
F-A P/S upper bound
realtime
1.8e+05 error
omission error
1.6e+05
1.4e+05
omission error
1.2e+05
time [us]
1.0e+05
8.0e+04
6.0e+04
4.0e+04
yes
2.0e+04 no
10 20 30 40 50 60 70 80 90 100
received publication #
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
42. Motivation Overview The Upper Bound Evaluation Summary 19 of 27 slides
Publication Transmission Time – Error (national)
107
F-A P/S upper bound
realtime
error
omission error
106
omission error
time [us]
105
yes
104 no
260 280 300 320 340 360 380 400
received publication #
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
43. Motivation Overview The Upper Bound Evaluation Summary 20 of 27 slides
Processing Time vs Transmission Time (national)
network transmission delay
processing time
107
106
time [us]
105
104
103
0 50 100 150 200 250 300
message
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
44. Motivation Overview The Upper Bound Evaluation Summary 21 of 27 slides
Upper Bound vs NTP (national)
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
45. Motivation Overview The Upper Bound Evaluation Summary 22 of 27 slides
Upper Bound vs NTP (global)
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
46. Motivation Overview The Upper Bound Evaluation Summary 23 of 27 slides
Upper Bound vs Real Time (national)
120
upper bound - realtime
100
80
number of messages
60
40
20
0
30000 35000 40000 45000 50000 55000 60000 65000 70000 75000 80000
error [us]
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
47. Motivation Overview The Upper Bound Evaluation Summary 24 of 27 slides
Upper Bound vs Real Time (global)
40
upper bound - realtime
35
30
number of messages
25
20
15
10
5
0
9.1e+05 9.3e+05 9.5e+05 9.7e+05 9.9e+05
error [us]
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
48. Motivation Overview The Upper Bound Evaluation Summary 25 of 27 slides
Summary & Outlook
Fail-Awareness in Publish/Subscribe
Better or comparable with NTP
No constraints on the system
Outlook
Take a closer look at omissions
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
49. Motivation Overview The Upper Bound Evaluation Summary 25 of 27 slides
Summary & Outlook
Fail-Awareness in Publish/Subscribe
Better or comparable with NTP
No constraints on the system
Outlook
Take a closer look at omissions
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
50. Motivation Overview The Upper Bound Evaluation Summary 26 of 27 slides
Thank You!
http://wwwse.inf.tu-dresden.de/zib/
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach
51. Motivation Overview The Upper Bound Evaluation Summary 27 of 27 slides
References
Antonio Carzaniga, David S. Rosenblum, and Alexander L. Wolf.
Design and evaluation of a wide-area event notification service.
ACM Trans. Comput. Syst., 19(3):332–383, 2001.
Christof Fetzer and Flaviu Cristian.
Fail-awareness: An approach to construct fail-safe applications.
Journal of Real-Time Systems, pages 203–238, March 2003.
Alan Mislove, Ansley Post, Andreas Haeberlen, and Peter Druschely.
Experiences in building and operating ePOST, a reliable peer-to-peer pplication.
In EuroSys, 2006.
Fail-Aware Publish/Subscribe Zbigniew Jerzak, Christof Fetzer, Robert Fach