Download - Fail-Aware Publish/Subscribe

Motivation Overview The Upper Bound Evaluation Summary 1 of 27 slides

Fail-Aware Publish/SubscribeNCA 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

mailto:[email protected]

State-of-the-art

I Problems:I we have conversationI we need dissemination

I Dissemination:I you are interested in data (not source)

I Hence loose-coupling:I anonymousI asynchronous

State-of-the-art

State-of-the-art

Content-Based Systems

I Peer-to-Peer, Publish/Subscribe, CBN...

I Content-/Name-drivenI Loosely-coupled and distributedI Failures are imminent [MPHD06]

Content-Based Systems

I Peer-to-Peer, Publish/Subscribe, CBN...

I Content-/Name-drivenI Loosely-coupled and distributedI Failures are imminent [MPHD06]

Fail-Awareness

...allows an application to [...] detect when it cannot guarantee allits safety properties and in particular, to detect when to switch toa fail-safe mode [FC03]

I Did we receive all messages?I Was the received message timely?

Fail-Awareness

...allows an application to [...] detect when it cannot guarantee allits safety properties and in particular, to detect when to switch toa fail-safe mode [FC03]

I Did we receive all messages?I Was the received message timely?

Publish/Subscribe Systems [CRW01]

Time in Publish/Subscribe

I no clock synchronisationI no violation of loose-couplingI many-to-many communication

How long did it take for a message to be delivered from p to q?

I no clock synchronisationI no violation of loose-couplingI many-to-many communication

How long did it take for a message to be delivered from p to q?

Idea

I do not calculate the transmission timeI calculate upper bound on transmission time

Message transmission time is no longer than the upper bound

Idea

I do not calculate the transmission timeI calculate upper bound on transmission time

Message transmission time is no longer than the upper bound

Upper Bound – Point to Point

Upper Bound – Publish/Subscribe

Upper Bound – Optimizations

Upper Bound vs NTP

NTP

I no guarantees whatsoeverI 2*max error gives estimated upper bound

Upper Bound

I preciseI more accurate

Experiment Setup

TSC as Time Source

0

10

20

30

40

50

60

01:04 01:08 01:12 01:16 01:20 02:00 02:04 02:08 02:12 02:16

drif

t rat

e[pp

m]

days:hours

planetlab-1a.ics.uci.eduplanetlab01.mpi-sws.mpg.de

planetlab-2.cse.ohio-state.eduplanetlab2.exp-math.uni-essen.de

TSC as Time Source – Single Host

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

01:00 01:12 02:00 02:12 03:00 03:12 04:00 04:12 05:00 05:12

drif

t rat

e[pp

m]

days:hours

planetlab-2.cse.ohio-state.eduerror

Publication Transmission Time (national)

2.0e+04

4.0e+04

6.0e+04

8.0e+04

1.0e+05

1.2e+05

1.4e+05

1.6e+05

1.8e+05

2.0e+05

0 500 1000 1500 2000 2500 3000noyes

time

[us]

omis

sion

err

or

received publication #

F-A P/S upper boundrealtime

erroromission error

Publication Transmission Time (global)

105

106

107

108

0 500 1000 1500 2000noyes

time

[us]

omis

sion

err

or

erroromission error

Publication Transmission Time – Bootup (national)

2.0e+04

4.0e+04

6.0e+04

8.0e+04

1.0e+05

1.2e+05

1.4e+05

1.6e+05

1.8e+05

2.0e+05

10 20 30 40 50 60 70 80 90 100noyes

time

[us]

omis

sion

err

or

erroromission error

Publication Transmission Time – Error (national)

104

105

106

107

260 280 300 320 340 360 380 400noyes

time

[us]

omis

sion

err

or

erroromission error

Processing Time vs Transmission Time (national)

103

104

105

106

107

0 50 100 150 200 250 300

time

[us]

message

network transmission delayprocessing time

Upper Bound vs NTP (national)

Upper Bound vs NTP (global)

Upper Bound vs Real Time (national)

0

20

40

60

80

100

120

30000 35000 40000 45000 50000 55000 60000 65000 70000 75000 80000

num

ber

of m

essa

ges

error [us]

upper bound - realtime

Upper Bound vs Real Time (global)

0

5

10

15

20

25

30

35

40

9.1e+05 9.3e+05 9.5e+05 9.7e+05 9.9e+05

num

ber

of m

essa

ges

error [us]

upper bound - realtime

Summary & Outlook

I Fail-Awareness in Publish/SubscribeI Better or comparable with NTPI No constraints on the system

Outlook

I Take a closer look at omissions

Summary & Outlook

I Fail-Awareness in Publish/SubscribeI Better or comparable with NTPI No constraints on the system

Outlook

I Take a closer look at omissions

Thank You!http://wwwse.inf.tu-dresden.de/zib/

http://wwwse.inf.tu-dresden.de/zib/

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.