Download DEPENDABLE SYSTEMS, NETWORKS AND SERVICES

National Aerospace University "KhAI“ (Ukraine)
Newcastle University (UK)
Yuhui Chen,
Anatoliy Gorbenko, Vyacheslav Kharchenko,
Alexander Romanovsky, Olga Tarasyuk
The Threat
of Uncertainty
in Service-Oriented
Architecture
1
OBJECTIVE
The objective of the study
is to investigate the uncertainty of response time
and performance of Web Services and instability
of a communication medium (the Internet) as well
as their influence on SOA dependability
OVERVIEW
Setting the Experiments
TCP Workflow Overview
Delay Analysis
Invocation Delay Analysis
PINGing Delay Analysis
Tracing Route Analysis
WS Performance Assessment
Summary
Setting the Experiments
Newcastle
University, UK
*.ncl.ac.uk
Web Server
DNA Databank, Japan
ping
Start time: Jun 04 20:01:24
(DDBJ)
ping
2 second)
(every
End time:
Jun
...09 08:02:51
Clients
http://xml.nig.ac.jp
Invoke Fasta
Total number of invokes:
> 650 Blast
Invoke
(every 10 minutes)
Total number
of pings:
ping
> 200000
...
KhAI University, Ukraine
*.khai.edu
Fasta
WS
Blast
WS
...
TCP Workflow Overview
RPT  RT – 2*RTT
RPT - Request Processing Time
RT - Response Time
RTT - Round Trip Time
Response Delay Analysis (1)
Invoking FASTA WS from NU
Fasta WS - ResponseTime
RT, ms
1900
1700
1500
1300
Time slot 2
1100
900
0
50
100
150
200
250 300 350
Invocation No
400
450
500
550
600
650
Response Delay Analysis (2)
Invoking FASTA WS from NU
0.6
Fasta WS - Probability distribution series of RT
0.5
0.4
0.3
0.2
0.1
RT, ms
0
<950
1050
1150
1250
1350
1450
1550
1650
1750
1850
1950 >1950
P 0.3492 0.562 0.0168 0.0107 0.0061 0.0107 0.0046 0.0077 0.0245 0.0031 0.0031 0.0015
Response Delay Analysis (3)
Invoking BLAST WS from NU
Blast WS - ResponseTime
RT, ms
1900
1700
1500
1300
Time slot 2
1100
900
0
50
100
150
200
250 300 350
Invocation No
400
450
500
550
600
650
Response Delay Analysis (4)
Invoking BLAST WS from NU
0.6
Blast WS - Probability distribution series of RT
0.5
0.4
0.3
0.2
0.1
RT, ms
0
<950
1050
1150
1250
1350
1450
1550
1650
1750
1850
1950 >1950
P 0.1776 0.5574 0.0123 0.0184 0.0904 0.0444 0.0276 0.0092 0.0444 0.0031 0.0077 0.0077
Response Delay Analysis (5)
Summary
Invocation response time (RT), ms
Time slot
min.
max.
Fasta WS
av.
std. dev.
Time slot 1
937
1953
996.91
163.28
Time slot 2
937
4703
Blast WS
1087.28
171.12
Time slot 1
1000
1015
1071.17
1265.72
291.57
572.70
Time slot 2
1750
3453
Deviation takes from 15 to 50% (!!!) of average RT
PINGing Delay Analysis (1)
Time Slots
Testing Interval
333ms
309ms
Stable
network
3
1
309ms
Stable
network
Jun 09 08:02:51
1
Jun 07 17:00:00
Jun 07 17:19:55
Stable
network
2
Jun 06 02:31:30
309ms
Jun 05 23:23:48
Jun 04 20:01:24
Time_slot_1
PINGing Delay Analysis (2)
Time Slot_1 (network delay is high stable)
PING from Newcastle University (Time_slot_1)
- Probability distribution series of RTT
0.8
0.6
Total duration was about 105 hours
0.4
Number of intermediate hosts was 17
0.2
RTT, ms
0
309
310
311
312
313
314
315
316
317
P 0.855 0.121 0.012 0.005 0.003 0.002 3E-04 5E-04 1E-04
318
0
319 >319
0
5E-04
PINGing Delay Analysis (3)
Time Slot_2 (network delay is stable enough)
0.6
PING from Newcastle University (Time_slot_2)
- Probability distribution series of RTT
0.5
0.4
The duration was about 3 hours
0.3
Number of intermediate hosts was 20
0.2
0.1
RTT, ms
0
332
333
334
335
336
337
338
339
340
P 0.362 0.604 0.023 0.006 0.003 0.002 4E-04 2E-04 9E-05
341
342 >342
0
9E-05 2E-04
PINGing Delay Analysis (4)
Time Slot_3 (network delay is unstable)
0.4
0.3
PING from Newcastle University
(Time_slot_3) - Probability
distribution series of RTT
The duration was about 20 min
Number of intermediate
hosts was 17 (= TimeSlot_1)
0.2
0.1
RTT, ms
0
309
310
311
312
313
314
315
316
317
318
319 >319
P 0.025 0.037 0.061 0.107 0.107 0.265 0.354 0.032 0.005 0.002 0.003 0.003
PINGing Delay Analysis (5)
PINGing DDBJ host from KhAI
0.4
PING - Probability distribution series of RTT
The duration was about 2 days
0.3
Number of intermediate hosts was 26
0.2
0.1
RTT, ms
0
<351 371
391
411
431
451
471
491
511
531
551 >551
P 0.395 0.165 0.077 0.08 0.069 0.058 0.04 0.029 0.032 0.021 0.011 0.023
PINGing Delay Analysis (6)
Summary
Ping’s round trip time (RTT), ms
Time slot
min.
max.
av.
std. dev.
PINGing from Newcastle University’s LAN (UK)
Time slot 1
309
422
309.21
1.40
Time slot 2
332
699
332.72
3.48
Time slot 3
309
735
312.94
12.73
PINGing from KhAI University’s LAN (Kharkiv, Ukraine)
-
341
994
396.27
62.14
Tracing Route Analysis
Route length
Newcastle
University, UK
*.ncl.ac.uk
17 routers
...
...
KhAI University,
Ukraine
*.khai.edu
26 intermediate hosts
(routers)
DNA Databank,
Japan (DDBJ)
http://xml.nig.ac.jp
TRACERT Delay Analysis (1)
Tracing Route from KhAI
min
Round Trip Time, ms
av
max
std dev
4
4
4
4
5
15
50
50
53
53
57
64
55
62
63
69
138
137
137
140
313
323
341
343
341
342
44.22
31.28
39.72
43.50
80.39
168.72
121.17
95.06
108.75
101.17
75.00
116.94
79.22
86.56
84.72
114.50
158.33
151.83
153.61
165.17
327.50
347.67
355.67
374.00
370.00
374.72
Min
4
Av
126
117
155
175
365
436
281
225
342
188
116
222
115
129
155
250
203
210
196
333
391
445
460
518
458
482
44.43
36.92
49.85
55.71
98.49
117.94
73.27
56.13
86.13
48.78
27.53
46.98
18.14
21.84
30.68
57.31
18.86
15.53
22.17
52.06
20.49
33.21
27.69
44.77
39.11
41.96
IP
10.3.128.1
80.249.231.121
217.112.212.69
80.249.224.55
80.249.224.97
80.91.177.85
80.91.160.206
217.28.250.41
212.162.25.5
4.68.118.94
4.69.132.126
4.69.132.137
4.69.132.130
4.69.133.89
4.69.133.86
4.69.132.133
4.69.137.74
4.69.134.74
4.68.16.142
4.78.132.18
150.99.20357
150.99.203.26
150.99.197.158
133.39.27.21
133.39.28.1
133.39.105.31
Max Std Dev
Intermediate host
DNS
proxy.khai.edu
121.231.249.80.customer.teleportsv.net
69.212.112.217.unknown.teleportsv.net
IPTN-SW02-1.teleportsv.net
IP-RT00.teleportsv.net
teleportsv.tr1-v180.ua-kiev.datagroup.ua
tr1-v454.de-fra.datagroup.ua
r9-ge-0-0-3-23-Fra-Anct.DE.DataBone.net
IP
DNS-name
44.22 126 44.43 10.3.128.1 proxy.khai.edu
ae-31-53.ebr1.Frankfurt1.Level3.net
ae-1-100.ebr2.Frankfurt1.Level3.net
ae-2.ebr1.Dusseldorf1.Level3.net
ae-1-100.ebr2.Dusseldorf1.Level3.net
ae-2.ebr1.Amsterdam1.Level3.net
ae-1-100.ebr2.Amsterdam1.Level3.net
ae-2.ebr2.London1.Level3.net
ae-43.ebr1.NewYork1.Level3.net
ae-81-81.csw3.NewYork1.Level3.net
ae-3-89.edge1.NewYork1.Level3.net
JAPAN-TELEC.edge1.NewYork1.Level3.net
tokyo1-dc-RM-P-2-3-0-11.sinet.ad.jp
nagoya-dc-RM-AE-0-11.sinet.ad.jp
nig-Lan.sinet.ad.jp
fwb-1.nig.ac.jp
oak.genes.nig.ac.jp
TRACERT Delay Analysis (2)
Tracing Route
DNA Databank,
Japan (DDBJ)
http://xml.nig.ac.jp
KhAI University,
Kharkiv, Ukraine
*.khai.edu
Ukraine
Kiev
Amsterdam
New-York
Holand
USA
Frankfurt
London
Tokio
Dusseldorf
Germany
Unstable Network
Nagoya
UK
Japan
High-Stable Network
Newcastle
University, UK
*.ncl.ac.uk
TRACERT Delay Analysis (3)
Tracing Route
Newcastle
Moscow
Peking
London
Frankfurt
Kharkiv
New-York
Tokio
WS Performance Assessment
Approximate estimation of Web Service’s
Request Processing Time (RPT) taking of
networks delay
Minimal RPT, ms
Fasta WS
Blast WS
Time_Slot_1
319
319.00
Time_Slot_2
336
351.00
divergence, %
5.06
9.12
RPT  RT – 2*RTT
RPT - Request Processing Time
RT - Response Time (Invocation)
RTT - Round Trip Time (Ping)
Summary
Deviation of Response Time (RT) takes from 15 to 50% (!!!)
of average one for Fasta and Blast WSs
It is noteworthy that even in spite of sufficiently stable
network delay during Time_slot_1 and Time_slot_2 a response
time of the Fasta and, especially, Blast WSs has significant
instability that can be explained only by internal reasons or
unstable WS loading.
6.3% (Blast) and 4% (Fasta) of the requests had the
response time 1.5 times greater than the average one, and for
several responses it took even 5 times greater. These cases
would potentially cause timing errors
Network brings additional uncertainty into response time
Discussion
Network instability significantly depends on the
QoS of a local Internet Service Provider (ISP) and
network route.
Occasional transient and long-term Internet
congestions, packet losses and network route
changes that are difficult-to-predict also reduce
stability of SOS operation.
Solutions
Good measurement of uncertainty is important
but this is only the beginning.
Uncertainty existing in SOA should be treated as a threat
to dependability (similar and in addition to the faults, errors
and failures).
This issue will require developing new resilience-explicit
techniques and end-to-end QoS mechanisms.
Solutions
The future solutions will need to deal with a number of
issues such as uncertainty of fault assumptions, uncertainty of
components behaviour and dependability, uncertainty of error
detection, etc.
One of the possible solutions for resisting the uncertainty is
to use service and path redundancy and diversity inherent to
SOA.
The traditional adaptive techniques based on the control
feedback will not be directly applicable in the current form as
they are intended for predictable behaviour.