Traffic Analysis and model synthesis of remote instrumentation - dorii
Traffic Analysis and model synthesis of remote instrumentation - dorii
Traffic Analysis and model synthesis of remote instrumentation - dorii
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
contract # 213110<br />
Deliverable DSA1.5 (Update)<br />
<strong>Traffic</strong> <strong>Analysis</strong> <strong>and</strong> <strong>model</strong> <strong>synthesis</strong> <strong>of</strong> <strong>remote</strong> <strong>instrumentation</strong><br />
Abstract<br />
This report updates DSA1.5 with further experimental results.<br />
workpackage SA1<br />
lead editor CNIT<br />
partners CNIT, GRNET, USTUTT<br />
document classification public
Delivery Slip<br />
Name Partner Date Signature<br />
Document Log<br />
# Date Summary <strong>of</strong> Changes Authors<br />
1 20/07/2010 Table <strong>of</strong> Contents Davide Adami<br />
2 27/07/2010 Section 4, Section 3 Davide Adami, Alexey Cheptsov, Giorgio<br />
Bolzon, Kasia Bylec<br />
3 30/07/2010 Section 5 Davide Adami, Matteo Lanati, Anastasios<br />
Zafeiropolous<br />
4 31/07/2010 Introduction, Conclusions,<br />
revision<br />
Franco Davoli<br />
DSA1.5 PUBLIC 2
Table <strong>of</strong> Contents<br />
1. INTRODUCTION...........................................................................................................................6<br />
2. NETWORK INFRASTRUCTURE ...................................................................................................7<br />
2.1. Best effort IPv4 ........................................................................................................................7<br />
2.2. IPv6 Testbed.............................................................................................................................8<br />
3. PERFORMANCE EVALUATION PROCEDURE .............................................................................9<br />
4. PERFORMANCE MONITORING.................................................................................................10<br />
4.1. EEWS .....................................................................................................................................10<br />
4.1.1. TEST DESCRIPTION ..............................................................................................................10<br />
4.1.2. EXPERIMENTAL RESULTS WITH A BEST EFFORT NETWORK.................................................11<br />
4.2. OPATM-BFM ........................................................................................................................17<br />
4.2.1. TEST DESCRIPTION ..............................................................................................................17<br />
4.2.2. EXPERIMENTAL RESULTS ....................................................................................................18<br />
5. IPV6 TESTBED DEPLOYMENT..................................................................................................22<br />
5.1.1. IPV6 TESTBED EXTENSION..................................................................................................22<br />
6. CONCLUSIONS ..........................................................................................................................25<br />
7. LIST OF ACRONYMS .................................................................................................................26<br />
REFERENCES....................................................................................................................................28<br />
CONTACT INFORMATION ................................................................................................................29<br />
DSA1.5 PUBLIC 3
Illustration Index<br />
Figure 1 DORII Network Infrastructure .............................................................................................7<br />
Figure 2 IPv6 Testbed.........................................................................................................................8<br />
Figure 3 Performance evaluation procedure.......................................................................................9<br />
Figure 4 Earthquake Early Warning System data storage <strong>and</strong> processing .......................................10<br />
Figure 5 EEWS Test Scenario – Block Diagram <strong>and</strong> network monitoring ......................................11<br />
Figure 6 Input/Output Rate at ie-01.eucentre.interface via SNMP monitoring ................................12<br />
Figure 7 Latency from EUCENTRE to se01.kallisto.hellasgrid.gr ..................................................12<br />
Figure 8 Latency from EUCENTRE to se01.ariagni.hellasgrid.gr...................................................12<br />
Figure 9 Latency from EUCENTRE to se01.marie.hellasgrid.gr.....................................................13<br />
Figure 10 Latency from EUCENTRE to se.reef.man.poznan.pl ......................................................13<br />
Figure 11 Pathload output: available b<strong>and</strong>width estimated from GRNET to EUCENTRE .............13<br />
Figure 12 Pathload output: available b<strong>and</strong>width estimated from PSNC to EUCENTRE.................14<br />
Figure 13 Pathload output: available b<strong>and</strong>width estimated from PSNC to GRNET ........................14<br />
Figure 14 Input <strong>and</strong> output rate (Byte/s) at se01.kallisto.hellasgrid.gr interface on July 6, 2010 ....16<br />
Figure 15 OPATM-BFM -Test Scenario: Block Diagram <strong>and</strong> Network Monitoring.......................18<br />
Figure 16 Latency between OGS <strong>and</strong> se01.kallisto.hellasgrid.it (week 24).....................................19<br />
Figure 17 Latency between PSNC <strong>and</strong> se01.kallisto.hellasgrid.it (week 25) ...................................19<br />
Figure 18 Input <strong>and</strong> output rate (Byte/s) at se01.kallisto.hellasgrid.gr interface on June 17, 2010..20<br />
Figure 19 Pathload output: available b<strong>and</strong>width from PSNC to GRNET on June 23, 2010 ............20<br />
Figure 20 Input <strong>and</strong> output rate (Byte/s) at se01.kallisto.hellasgrid.gr interface on June 23, 2010..21<br />
Figure 21 <strong>Traffic</strong> Volume in Bit/s between Hellasgrid <strong>and</strong> OGS (week 25)....................................21<br />
Figure 22 <strong>Traffic</strong> Volume in Bit/s between Hellasgrid <strong>and</strong> PSNC during week 25.........................21<br />
Figure 23 Test bed Setup ..................................................................................................................23<br />
Figure 24 IPv6 <strong>Traffic</strong> Monitoring ...................................................................................................24<br />
Figure 25 IPv6 <strong>Traffic</strong> traces between the VCR <strong>and</strong> the IE .............................................................24<br />
DSA1.5 PUBLIC 4
Executive Summary<br />
The content <strong>of</strong> the present document is aimed at updating deliverable DSA1.5 “<strong>Traffic</strong> <strong>Analysis</strong><br />
<strong>and</strong> <strong>model</strong> <strong>synthesis</strong> <strong>of</strong> <strong>remote</strong> <strong>instrumentation</strong>” with the results <strong>of</strong> further experiments <strong>and</strong><br />
measurement campaign carried out during the last three months <strong>of</strong> the project.<br />
With respect to the previous release <strong>of</strong> DSA1.5, some relevant results have been added, especially<br />
as far as the IPv6 deployment is concerned. Besides the set-up <strong>of</strong> the IPv6 test bed between<br />
EUCENTRE <strong>and</strong> GRNET, two fundamental components developed in DORII (VCR <strong>and</strong> IE) have<br />
been ported to IPv6 <strong>and</strong> the Earthquake Early Warning System (EEWS) application from<br />
EUCENTRE is now IPv6-enabled for all communications between the VCR <strong>and</strong> the IE. Therefore,<br />
if the end-used is connected to a network <strong>of</strong>fering IPv6 support, other than the traditional IPv4, he<br />
can choose between IPv4 <strong>and</strong> IPv6 when connecting to that VCR, since IPV6 is fully supported as<br />
for the tools (VCR) <strong>and</strong> middleware (IE) developed in DORII. Unfortunately, since gLite does not<br />
fully support IPv6, communications with CE <strong>and</strong> SE are only possible via IPv4. The same<br />
consideration applies for the certification procedure.<br />
Finally, further tests have been carried out for EEWS <strong>and</strong> OPTAM-BFM applications.<br />
DSA1.5 PUBLIC 5
1. Introduction<br />
The deployment <strong>of</strong> DORII applications in their final form has allowed conducting more significant<br />
tests <strong>and</strong> evaluations also from the point <strong>of</strong> view <strong>of</strong> the networking infrastructure. In particular, the<br />
implementations <strong>of</strong> an IPv6-compliant VCR version <strong>and</strong> <strong>of</strong> an IPv6-compliant IE for the EEWS<br />
application <strong>of</strong> EUCENTRE have made possible to conduct connectivity <strong>and</strong> functionality tests on<br />
an IPv6 enabled network portion between EUCENTRE <strong>and</strong> HellasGrid sites at GRNET.<br />
At the same time, further network performance measurements have been conducted under the IPv4<br />
best effort service, still on EEWS <strong>and</strong> on OPATM-BFM.<br />
The present document reports the results <strong>of</strong> these tests, <strong>and</strong> it is organized as follows. We describe<br />
the updated networking infrastructure in Section 2, <strong>and</strong> briefly recall the performance evaluation<br />
methodology in Section 3. New experimental evaluation results for the two previously mentioned<br />
applications are reported <strong>and</strong> discussed in Section 4, whereas Section 5 highlights the IPv6 test bed<br />
<strong>and</strong> measurements. Section 6 contains the conclusions.<br />
DSA1.5 PUBLIC 6
Vacuum<br />
slits<br />
Monochromator<br />
Beam stopper<br />
Al filters<br />
SS<br />
Air Slits<br />
SETUP<br />
SYRM EP<br />
Fast<br />
Mammographic station<br />
shutters<br />
Ionization<br />
chambers<br />
BEAM<br />
2. Network Infrastructure<br />
Two different network scenarios are taken into account:<br />
• IPv 4 best effort network infrastructure<br />
• Best effort network infrastructure with IPv6 Support<br />
A short description is provided in the following sub-sections.<br />
2.1. Best effort IPv4<br />
The basic DORII network infrastructure is reported in Figure 1 <strong>and</strong> has been extensively discussed<br />
in [1], [2]. Since the network infrastructure is used as is, i.e., in an “Internet-like” fashion, only a<br />
best effort packet delivery service is provided.<br />
© YSI<br />
© YSI<br />
Figure 1 DORII Network Infrastructure<br />
Figure 1 also highlights the location <strong>of</strong> the clients <strong>and</strong> servers deployed to monitor the available<br />
b<strong>and</strong>width <strong>and</strong> the Round-Trip Time (RTT), by using Pathload <strong>and</strong> Smokeping, respectively.<br />
DSA1.5 PUBLIC 7
2.2. IPv6 Testbed<br />
Figure 2 shows the IPv6 test bed that has been setup in the DORII project.<br />
LMU<br />
USTUTT<br />
DFN<br />
PIONIER<br />
(IPv4)<br />
PSNC<br />
UC<br />
RedIRIS<br />
GEANT<br />
(IPv6 Transport Network)<br />
GRNET<br />
(IPv6)<br />
CSIC<br />
IPv4<br />
Isl<strong>and</strong><br />
EUCENTRE<br />
(IPv6)<br />
OGS<br />
GARR<br />
(IPv6)<br />
CNIT<br />
ELETTRA<br />
IPv4<br />
Isl<strong>and</strong><br />
Native IPv6<br />
Figure 2 IPv6 Testbed<br />
Within the DORII network native IPv6 connectivity is provided from GRNET <strong>and</strong> PSNC through<br />
the GRNET <strong>and</strong> PIONEER networks that are connected through the GÉANT2 network.<br />
Furthermore, an IPv6 Test bed is set-up within EUCENTRE <strong>and</strong> a HellasGrid site in GRNET.<br />
Native IPv6 connectivity has been enabled in the access network for EUCENTRE <strong>and</strong><br />
ariagni.hellasgrid.gr that is a HellasGrid site in Crete. In EUCENTRE, IPv6 is supported in the<br />
EUCENTRE LAN, the hosting IE <strong>and</strong> the VCR, while in ariagni.hellasgrid.gr IPv6 is supported in<br />
all the hosts <strong>of</strong> the Grid site. IPv6 connection between EUCENTRE <strong>and</strong> ariagni.hellasgrid.gr is<br />
provided through the native IPv6 network <strong>of</strong> GRNET, GÉANT <strong>and</strong> GARR. The topology <strong>of</strong> the<br />
connection is shown in Figure 2.<br />
The EUCENTRE application is written in Java, <strong>and</strong> thus IPv6 is transparently supported for this<br />
application. However, since the gLite middleware is only partially ported to IPv6 at this moment,<br />
we were unable to execute tests with IPv6 to the HellasGrid sites. Thus, in order to make some<br />
pilot tests with IPv6 support, by using the DORII developed middleware, a special IPv6 enabled<br />
test bed is being created. In this test bed, an IPv6 enabled VCR is installed in GRNET in order to<br />
be accessible via IPv6 from the IE in EUCENTRE.<br />
More specifically, native IPv6 connectivity has been enabled between EUCENTRE <strong>and</strong> GRNET:<br />
• EUCENTRE LAN, hosting the IE<br />
• Italian NREN (GARR), GÉANT2<br />
• Greek NREN (GRNET), hosting VCR (IPv6-enabled), SEs <strong>and</strong> CEs<br />
DSA1.5 PUBLIC 8
3. Performance Evaluation Procedure<br />
As discussed in [3], to evaluate how the behavior <strong>of</strong> DORII applications is affected by the network<br />
performance, the evaluation procedure illustrated in Figure 3 has been defined.<br />
The basic idea behind our approach is to infer the impact <strong>of</strong> the network on the performance <strong>of</strong> the<br />
applications by measuring QoS metrics at the network layer <strong>and</strong> correlating them with metrics<br />
observed at the application level.<br />
Test<br />
Definition<br />
Test<br />
Running<br />
QoE<br />
Evaluation<br />
Statistics<br />
Statistics<br />
vs. QoE<br />
Network Parameter<br />
Configuration<br />
Guidelines<br />
Figure 3 Performance evaluation procedure<br />
More specifically, while running each application under test, network statistics are collected:<br />
indeed, the network monitoring platform provides delay <strong>and</strong> b<strong>and</strong>width measurements between<br />
specific couples <strong>of</strong> DORII sites, whereas the IEs, CEs <strong>and</strong> SEs are monitored by means <strong>of</strong> SNMP.<br />
At the application level, for each step, the end-user has to measure the performance metrics (e.g.,<br />
the overall execution time) that have to be optimized to improve the QoE from the user’s<br />
perspective.<br />
DSA1.5 PUBLIC 9
4. Performance Monitoring<br />
4.1. EEWS<br />
The EEWS application aims at recording seismic data from sensors, possibly in real time, <strong>and</strong> at<br />
processing them in order to extract time history for ground speed, ground acceleration <strong>and</strong><br />
displacements, the amplitude spectrum <strong>and</strong> the response spectrum. All the operations are performed<br />
<strong>remote</strong>ly <strong>and</strong> on the grid, employing the DORII infrastructure. The VCR plays the role <strong>of</strong> the user<br />
interface: all the actions are performed <strong>and</strong> all the resources are accessed through this web portal.<br />
Figure 4 Earthquake Early Warning System data storage <strong>and</strong> processing<br />
The IE is located at EUCENTRE (Pavia, Italy) <strong>and</strong> it hosts the IM devoted to access the server<br />
collecting data from sensors (by means <strong>of</strong> the Java client library). This node is in Genoa, Italy,<br />
while the seismic sensors are spread over the Liguria Region. Measurements from each channel are<br />
time-stamped <strong>and</strong> saved locally on the IE in a separate file, then moved to a SE. Finally, a CE<br />
retrieves each file from the SE <strong>and</strong> performs the computation. This task is carried out on a single<br />
binary, statically compiled <strong>and</strong> specified in the Input S<strong>and</strong>box. The user is asked only to select the<br />
input folder on the right SE. The output is downloaded to the home folder on the VCR. An<br />
alternative approach is represented by the Workflow Manager, a graphical <strong>and</strong> friendly interface to<br />
specify the parameters.<br />
4.1.1. Test Description<br />
The test scenario is depicted in Figure 5. Several tests have been performed by selecting different<br />
SEs <strong>and</strong> CEs.<br />
DSA1.5 PUBLIC 10
•1: Input Data Transfer<br />
•2: Data Retrieval<br />
•3: Data Processing<br />
•4: Output Data Transfer<br />
SE<br />
Seismic<br />
Sensor Network<br />
IE/IM<br />
Ie-01.unicentre.it<br />
1<br />
(B, D)<br />
(B, D)<br />
2<br />
4<br />
EUCENTRE<br />
(4)<br />
SE<br />
Network Metrics<br />
B = B<strong>and</strong>width<br />
D = Delay<br />
CE, SE, IE: SNMP-enabled interfaces<br />
3<br />
CE<br />
Figure 5 EEWS Test Scenario – Block Diagram <strong>and</strong> network monitoring<br />
4.1.2. Experimental Results with a best effort network<br />
The sequence <strong>of</strong> operations illustrated in Figure 5 represents the starting point for the test bed setup.<br />
Moreover, the performance <strong>of</strong> the network interconnecting the grid infrastructure is monitored<br />
during the entire life cycle <strong>of</strong> the application execution. Figure 5 also shows the network<br />
parameters monitored for each network path between couples <strong>of</strong> DORII grid resources. In the<br />
experiments, we skipped the acquisition phase, since the server gathering seismic sensors data is<br />
not part <strong>of</strong> the infrastructure, unlike the IE sending the query. Moreover, the required network<br />
resources are very limited: as a matter <strong>of</strong> fact, a single station channel only needs few kbps.<br />
The size <strong>of</strong> the file containing the initial data set for the computation is about 203 MB <strong>and</strong><br />
corresponds to measurement data coming from two channels <strong>and</strong> collected for a whole day. The<br />
archive is stored on a server at EUCENTRE that acts as IE <strong>and</strong> VCR; then, it is transferred to<br />
different SEs at GRNET <strong>and</strong> PSNC. The average throughput varies from 1.2 MB/s to 2.2 MB/s, the<br />
data transfer time goes from a minimum <strong>of</strong> 96 s to a maximum <strong>of</strong> 173 s, whereas the latency is low<br />
ranging from 30 ms to 55 ms (see Table 1 <strong>and</strong> Figure 7, Figure 8, Figure 9, Figure 10). The<br />
achieved results highlight that, in this experiment, the transfer time to the SE at PSNC is always<br />
less than the transfer times from EUCENTRE to different SEs located at GRNET.<br />
SE<br />
Upload<br />
Start Time<br />
Upload<br />
Time [s]<br />
Throughput<br />
[Mb/s]<br />
Average<br />
Latency<br />
[ms]<br />
se01.kallisto.hellasgrid.gr Jul 06, 12:01: 2010 165 1.25 53<br />
se01.ariagni.hellasgrid.gr Jul 06, 12:18: 2010 173 1.2 55<br />
se01.marie.hellasgrid.gr Jul 06, 12:23: 2010 156 1.3 50<br />
se01.reef.man.poznan.pl Jul 06, 12:27: 2010 96 2.2 30<br />
Table 1 Upload time from ie-01.eucentre.it to each SE (File Size: 203 MB)<br />
DSA1.5 PUBLIC 11
Figure 6 shows the input/output traffic rate measured by using SNMP-based monitoring tools at ie-<br />
01.eucentre,it (daily graph, 5 minutes average). It is easy to identify the traffic generated by<br />
moving the input file from the IE to the SEs (outbound direction). It is relevant to highlight that the<br />
max peak rate measured by SNMP is 949.3 kB/s, owing to the fact that SNMP computes the<br />
average over 5 minutes, whereas the file transfer length is always less than 5 minutes.<br />
Figure 6 Input/Output Rate at ie-01.eucentre.interface via SNMP monitoring<br />
Figure 7 Latency from EUCENTRE to se01.kallisto.hellasgrid.gr<br />
Figure 8 Latency from EUCENTRE to se01.ariagni.hellasgrid.gr<br />
DSA1.5 PUBLIC 12
Figure 9 Latency from EUCENTRE to se01.marie.hellasgrid.gr<br />
Figure 10 Latency from EUCENTRE to se.reef.man.poznan.pl<br />
Figure 11 Pathload output: available b<strong>and</strong>width estimated from GRNET to EUCENTRE<br />
DSA1.5 PUBLIC 13
Figure 12 Pathload output: available b<strong>and</strong>width estimated from PSNC to EUCENTRE<br />
The analysis <strong>of</strong> the data collected by the seismic sensor network is carried out by a parametric job,<br />
where the parameters correspond to different settings for the two seismic stations being monitored,<br />
so a single execution <strong>of</strong> the application produces two children nodes. The job is launched two<br />
times, <strong>and</strong> the target CEs are located at different sites: GRNET (ce01.athena.hellasgrid.gr), PSNC<br />
(ce.reef.man.poznan.pl) <strong>and</strong> IFCA-CSIC (egeece01.ifca.es).<br />
Table 2 reports the time necessary to upload the file from a specific SE (se01.kallisto.hellasgrid.gr)<br />
to each CE. As expected, the upload time is really low <strong>and</strong> the throughput is very high when we<br />
choose a CE located in the same network as the SE, whereas we have the highest upload time when<br />
the CE is located at IFCA-CSIC. It is relevant to highlight that although the available b<strong>and</strong>width<br />
between PSNC <strong>and</strong> GRNET estimated by using Pathload is more than 100 Mb/s (see Figure 13) ,<br />
when the selected CE is located at PSNC (ce.reef.man.poznan.pl) the throughput is less than 45<br />
Mb/s.<br />
CE<br />
Job Upload Time Throughput<br />
Submission Time [s]<br />
[MB/s]<br />
ce01.athena.hellasgrid.gr Jul 6 2010 12:45 6 35.8<br />
ce01.athena.hellasgrid.gr Jul 6 2010 12:45 8 24.1<br />
egeece01.ifca.es Jul 6 2010 12:46 99 2.1<br />
egeece01.ifca.es Jul 6 2010 12:46 87 2.4<br />
ce.reef.man.poznan.pl Jul 7 2010 16:57 36 5.5<br />
ce.reef.man.poznan.pl Jul 7 2010 16:57 43 4.9<br />
Table 2 EEWS Upload Time from se01.kallisto.hellasgrid.gr<br />
Figure 13 Pathload output: available b<strong>and</strong>width estimated from PSNC to GRNET<br />
DSA1.5 PUBLIC 14
Table 3 reports the processing time for each CE <strong>and</strong> the time necessary to download the output file<br />
to the selected SE (in this experiment, se01.kallisto.hellasgrid.gr again).<br />
Start<br />
CE<br />
Computation<br />
Time<br />
ce01.athena.hellasgrid.gr Jul 6 2010<br />
12:45:49<br />
ce01.athena.hellasgrid.gr Jul 6 2010<br />
12:45:50<br />
egeece01.ifca.es Jul 6 2010<br />
12:48:29<br />
egeece01.ifca.es Jul 6 2010<br />
12:48:20<br />
ce.reef.man.poznan.pl Jul 7 2010<br />
Processing<br />
Time [s]<br />
Start Getting<br />
Output<br />
Download<br />
Time [s]<br />
Throughput<br />
[MB/s]<br />
87 s Jul 6 2010<br />
12:47:16<br />
3 83.0<br />
7 s Jul 6 2010 8 17.9<br />
12:45:57<br />
62 s Jul 6 2010 22 9.8<br />
12:49:41<br />
4 s Jul 6 2010 36 5.9<br />
12:48:24<br />
62 Jul 7 2010 13 16.4<br />
16:58:02<br />
16:59:04<br />
4 Jul 7 2010 19 11.4<br />
16:58:13<br />
Table 3 Processing Time <strong>and</strong> Output Data Download Time<br />
ce.reef.man.poznan.pl Jul 7 2010<br />
16:58:09<br />
As clearly shown in the table, the processing time depends on the complexity <strong>of</strong> the job <strong>and</strong> on the<br />
processing power <strong>of</strong> the CE. In all the experiments, the time required for the first job to be<br />
processed (62 s -87 s) is greater than the time necessary for the second job (4 s - 7 s).<br />
The output files are retrieved, by using the VCR at EUCENTRE, which allows to control the<br />
movement <strong>of</strong> the files <strong>and</strong> to select the SE. Table 3 reports the time necessary to download the<br />
output files temporarily stored by each CE to se01.kallisto.hellasgrid.gr.<br />
Finally, Table 4 reports the overall time for the execution <strong>of</strong> the application. As clearly shown in the<br />
table, the amount <strong>of</strong> time necessary for exchanging the data may significantly affect the<br />
performance <strong>of</strong> the application <strong>and</strong> represents (except for the first <strong>and</strong> third case) the major<br />
component <strong>of</strong> the overall execution time.<br />
CE<br />
Total Time Processing Communication<br />
[s]<br />
Time [s] Time [s]<br />
ce01.athena.hellasgrid.gr 96 87 9<br />
ce01.athena.hellasgrid.gr 23 7 16<br />
ce.reef.man.poznan.pl 111 62 49<br />
ce.reef.man.poznan.pl 66 4 62<br />
egeece01.ifca.es 183 62 121<br />
egeece01.ifca.es 127 4 123<br />
Table 4 Total Time4<br />
As previously highlighted, the network-monitoring platform allows collecting information about<br />
the status <strong>of</strong> the network while the application is running.<br />
Errore. L'origine riferimento non è stata trovata., Figure 14 <strong>and</strong> Errore. L'origine riferimento<br />
non è stata trovata. show the input <strong>and</strong> output rate measured on July 6, 2010 at<br />
se01.kallisto.hellasgrid.gr by using SNMP.<br />
DSA1.5 PUBLIC 15
"Daily" Graph (5 Minute Average)<br />
Figure 14 Input <strong>and</strong> output rate (Byte/s) at se01.kallisto.hellasgrid.gr interface on July 6, 2010<br />
In summary, the upload <strong>and</strong> download time heavily affect the performance <strong>of</strong> the application <strong>and</strong>,<br />
in most cases, they represent the main components <strong>of</strong> the overall execution time. High data transfer<br />
time <strong>and</strong> low average throughput are achieved although the network is not congested, <strong>and</strong> this is<br />
probably due to the systems (WfMS, CE, SE) load <strong>and</strong> the protocol architecture.<br />
DSA1.5 PUBLIC 16
4.2. OPATM-BFM<br />
4.2.1. Test Description<br />
This application deals with the users belonging to the oceanographic <strong>model</strong>ing community: they<br />
work with numerical <strong>model</strong>s <strong>and</strong> with their data to get information about the past, the present or the<br />
future state <strong>of</strong> the marine environment or to study a specific process over limited space or time<br />
scales. The test consists <strong>of</strong> the following steps:<br />
1. Registration: the user has to apply to INFN for a personal certification.<br />
2. Login: the user opens the DORII VCR page hosted by OGS <strong>and</strong> asks for an account.<br />
3. Credential Management: once the user has both a valid certificate <strong>and</strong> a valid VCR<br />
account, the user can upload the public <strong>and</strong> the private key to the VCR in order to create a<br />
valid proxy at each connection.<br />
4. The workflow (from melisa.man.poznan.pl) may launch the simulation <strong>model</strong> with either<br />
(a) the most recent data or (b) historical data already stored on a SE.<br />
5. Logout<br />
a) In this case, the workflow asks the user to select the SE <strong>and</strong> the path where input<br />
data have to be uploaded from the IE (eva.ogs.trieste.it), the CE where simulations<br />
have to be run <strong>and</strong> finally the SE <strong>and</strong> the path where output data have to be<br />
downloaded. To optimize the performance <strong>of</strong> the application it is necessary to<br />
minimize the overall execution time. Therefore, from the end-user’s perspective,<br />
the following metrics have to be measured:<br />
• Data set upload time from IE to SE<br />
• Data set download time from SE to CE<br />
• Run time at CE<br />
• Data set upload time from CE to SE<br />
b) In this case, the workflow asks the user to select the SE <strong>and</strong> the path where input<br />
data are located, the CE where simulations have to be run <strong>and</strong> finally the SE <strong>and</strong><br />
the path where output data have to be downloaded. To optimize the performance <strong>of</strong><br />
the application it is necessary to minimize the overall execution time. Therefore,<br />
from the end-user’s perspective, the following metrics have to be measured:<br />
• Data set download time from SE to CE<br />
• Run time at CE<br />
• Data set upload time from CE to SE<br />
DSA1.5 PUBLIC 17
•1: Input Data Transfer<br />
•2: Data Retrieval<br />
•3: Data Processing<br />
•4: Output Data Transfer<br />
CINECA<br />
Repository<br />
The logic diagram depicted in<br />
GRNET<br />
IE/IM<br />
1<br />
eva.ogs.trieste.it<br />
(B, D)<br />
SE<br />
se01.kallisto.hellasgrid.gr<br />
(B, D)<br />
2<br />
4<br />
CINECA<br />
OGS<br />
(4)<br />
Network Metrics<br />
B = B<strong>and</strong>width<br />
D = Delay<br />
CE, SE, IE: SNMP-enabled interfaces<br />
PSNC<br />
3<br />
CE<br />
ce.reef.man.poznan.pl<br />
Figure 15 shows how data are transferred among the grid resources used by OPATM-BFM for<br />
retrieving the input data, running the simulations <strong>and</strong> archiving the output data.<br />
•1: Input Data Transfer<br />
•2: Data Retrieval<br />
•3: Data Processing<br />
•4: Output Data Transfer<br />
CINECA<br />
Repository<br />
IE/IM<br />
eva.ogs.trieste.it<br />
GRNET<br />
1<br />
(B, D)<br />
SE<br />
se01.kallisto.hellasgrid.gr<br />
(B, D)<br />
2<br />
4<br />
CINECA<br />
OGS<br />
(4)<br />
Network Metrics<br />
B = B<strong>and</strong>width<br />
D = Delay<br />
CE, SE, IE: SNMP-enabled interfaces<br />
PSNC<br />
3<br />
CE<br />
ce.reef.man.poznan.pl<br />
DSA1.5 PUBLIC 18
Figure 15 OPATM-BFM -Test Scenario: Block Diagram <strong>and</strong> Network Monitoring<br />
4.2.2. Experimental Results<br />
This sub-section reports the results <strong>of</strong> experimental tests carried out with the OPATM-BFM<br />
application. The time necessary to perform each operation detailed in<br />
•1: Input Data Transfer<br />
•2: Data Retrieval<br />
•3: Data Processing<br />
•4: Output Data Transfer<br />
CINECA<br />
Repository<br />
IE/IM<br />
eva.ogs.trieste.it<br />
GRNET<br />
1<br />
(B, D)<br />
SE<br />
se01.kallisto.hellasgrid.gr<br />
(B, D)<br />
2<br />
4<br />
CINECA<br />
Network Metrics<br />
B = B<strong>and</strong>width<br />
D = Delay<br />
CE, SE, IE: SNMP-enabled interfaces<br />
OGS<br />
PSNC<br />
3<br />
(4)<br />
CE<br />
ce.reef.man.poznan.pl<br />
Figure 15 has been measured at the application level.<br />
In this test, input data are initially located at OGS (adamo.ogs.trieste.it) <strong>and</strong> are transferred to a SE<br />
(se01.kallisto.hellasgrid.gr) located at GRNET.<br />
a) from IE (eva.ogs.trieste.it) to SE (se01.kallisto.hellasgrid.gr). Input Data: 8 files, 8 GB<br />
totally). The following table reports the data transfer time, whereas Figure 16 shows the<br />
latency measured between OGS <strong>and</strong> se01.kallisto.hellasgrid.gr over one month. When the<br />
data transfer was performed the latency was less than 100 ms, but highly varying<br />
SE Start Upload Time Data Transfer Time<br />
se01.kallisto.hellasgrid.gr Jun 17 2010 10:30 892 s (8.9 MB/s)<br />
DSA1.5 PUBLIC 19
Figure 16 Latency between OGS <strong>and</strong> se01.kallisto.hellasgrid.it (week 24)<br />
b) from SE (se01.kallisto.hellasgrid.gr) to CE (ce.reef.man.poznan.pl). Input Data: 1 File, 940 MB.<br />
The following table reports the data transfer time, whereas Figure 17 shows the latency<br />
measured between PSNC <strong>and</strong> se01.kallisto.hellasgrid.gr over one month. When the data<br />
transfer was performed the latency was around 65 ms.<br />
CE Start Upload Time Data Transfer Time<br />
ce.reef.man.poznan.pl Jun 23 2010 10:30 159 s (5.9 MB/s)<br />
Figure 17 Latency between PSNC <strong>and</strong> se01.kallisto.hellasgrid.it (week 25)<br />
c) The following table reports CE Processing Time (ce.reef.man.poznan.pl).<br />
CE Start Computation Time Run Time<br />
ce.reef.man.poznan.pl Jun 23 2010 10:33 360 s<br />
d) from CE (ce.reef.man.poznan.pl) to SE (se01.kallisto.hellasgrid.gr) Output Data: 1 File, 4.6 GB.<br />
The following table reports the data transfer time,<br />
DSA1.5 PUBLIC 20
SE Start Upload Time Data Transfer Time<br />
se01.kallisto.hellasgrid.gr Jun 23 2010 10:40 1240 s (3.7 MB/s)<br />
The total execution time was 1513 s. Since the computation time was 360 s), the main percentage<br />
<strong>of</strong> the total execution time (76%) is due to the data transfer time <strong>and</strong> therefore it is necessary to<br />
investigate how the network b<strong>and</strong>width is used by the application.<br />
Figure 18 highlights the throughput measured at the interface <strong>of</strong> kallisto.hellasgrid.gr on June 17,<br />
2010 (step a). As shown in the graph, the maximum inbound throughput was 9.69 MB/s.<br />
Figure 18 Input <strong>and</strong> output rate (Byte/s) at se01.kallisto.hellasgrid.gr interface on June 17, 2010<br />
Figure 19 shows the available b<strong>and</strong>width from PSNC to GRNET estimated by using pathload on<br />
June 23, 2010 <strong>and</strong> highlights that the during the data transfers (step b <strong>and</strong> step d) the available<br />
b<strong>and</strong>width is not completely used.<br />
Figure 19 Pathload output: available b<strong>and</strong>width from PSNC to GRNET on June 23, 2010<br />
Figure 20 illustrates the throughput, measured by using SNMP, at se01.kallisto.hellasgrid.gr<br />
interface on June 23, 2010. Both data transfers (step b <strong>and</strong> d) are highlighted.<br />
DSA1.5 PUBLIC 21
Figure 20 Input <strong>and</strong> output rate (Byte/s) at se01.kallisto.hellasgrid.gr interface on June 23, 2010<br />
Finally, Figure 21 <strong>and</strong> Figure 22 shows the traffic volume measured between Hellasgrid <strong>and</strong> OGS<br />
(week 25) as well as between Hellasgrid <strong>and</strong> PSNC (see Wednesday 23 June, 2010 when the<br />
experimental test was performed.<br />
Figure 21 <strong>Traffic</strong> Volume in Bit/s between Hellasgrid <strong>and</strong> OGS (week 25)<br />
Figure 22 <strong>Traffic</strong> Volume in Bit/s between Hellasgrid <strong>and</strong> PSNC during week 25<br />
DSA1.5 PUBLIC 22
5. IPv6 Testbed Deployment<br />
5.1.1. IPv6 Testbed Extension<br />
This section reports the results <strong>of</strong> some “pilot” tests that were carried out to demonstrate IPv6<br />
support in the VCR <strong>and</strong> IE components. These tests involved EUCENTRE <strong>and</strong> GRNET <strong>and</strong> the<br />
network infrastructure is shown in Figure 2. It is important to note that EUCENTRE’s EEWS<br />
application is written in Java <strong>and</strong> thus IPv6 is transparently supported by this application. However,<br />
since the gLite middleware is partially ported to IPv6 at the time <strong>of</strong> writing, we were unable to<br />
execute tests with IPv6 on the HellasGrid sites. Thus, in order to make some pilot tests with IPv6<br />
support, by using the DORII developed middleware, a special IPv6 enabled test bed was created<br />
<strong>and</strong> an-hoc IPv6-enabled VCR installed.<br />
The testing scenario includes the above mentioned components (VCR, IE), but EUCENTRE’s<br />
EEWS application workflow was executed without the integration <strong>of</strong> the gLite security<br />
components, which -at the time <strong>of</strong> executing the experiments- are not yet IPv6 enabled. It is<br />
envisaged that when the gLite security components will support IPv6, the final integration with the<br />
testing setup will be easy.<br />
In the test bed, a virtual machine is set up at GRNET where the VCR for the EUCENTRE<br />
application is hosted. The virtual machine is IPv6 enabled with the following network setup:<br />
• Hostname: <strong>dorii</strong>-vcr-v6.admin.grnet.gr<br />
• IPv4 address: 195.251.28.138<br />
• IPv6 address: 2001:648:2320:7:250:56ff:fe9b:4806<br />
On the EUCENTRE site the IE is set up in an IPv6 enabled machine <strong>and</strong> network, with the<br />
following network setup:<br />
• Hostname: ie-01.eucentre.it<br />
• IPv4 address: 193.296.66.121<br />
• IPv6 address: 2001:760:2000:2000::28<br />
A traceroute comm<strong>and</strong> between these sites was launched to check IPv6 reachability, as shown in<br />
the following:<br />
[root@<strong>dorii</strong>-vcr-v6 ~]# traceroute 2001:760:2000:2000::28<br />
traceroute to 2001:760:2000:2000::28 (2001:760:2000:2000::28), 30 hops max, 40 byte packets<br />
1 2001:648:2320:7::2 (2001:648:2320:7::2) 0.932 ms 0.952 ms 0.986 ms<br />
2 grnetrouter.grnet-admin.athens-3.access-link.grnet.gr (2001:648:2ffd:f057:1::1) 2.561 ms * *<br />
3 eie2-to-ath3.backbone.grnet.gr (2001:648:2fff:1103::1) 92.446 ms 188.988 ms 188.954 ms<br />
4 grnet.rt1.ath2.gr.geant2.net (2001:798:19:10aa::1) 163.544 ms 163.595 ms 163.693 ms<br />
5 as0.rt1.vie.at.geant2.net (2001:798:cc:1001:1901::5) 171.833 ms 172.003 ms 172.042 ms<br />
6 so-3-0-0.rt1.mil.it.geant2.net (2001:798:cc:1001:1e01::2) 184.599 ms 179.513 ms 176.494 ms<br />
7 garr-gw.rt1.mil.it.geant2.net (2001:798:1e:10aa::2) 173.280 ms 173.424 ms 93.156 ms<br />
8 rt1-mi1-rt-mi3.mi3.garr.net (2001:760:ffff:ffff::f7) 45.704 ms 45.758 ms 45.731 ms<br />
9 rt-mi3-rc-pv.pv.garr.net (2001:760:ffff:ffff::e5) 46.521 ms 46.470 ms 46.601 ms<br />
10 rc-pv-ru-unipv.pv.garr.net (2001:760:ffff:16c::41) 47.184 ms 47.211 ms 47.161 ms<br />
DSA1.5 PUBLIC 23
11 2001:760:2000::8 (2001:760:2000::8) 47.167 ms 47.023 ms 46.995 ms<br />
12 2001:760:2000:2000::28 (2001:760:2000:2000::28) 46.591 ms 46.575 ms 46.534 ms<br />
Browser<br />
Instrument element<br />
(EUCENTRE)<br />
2001:760:2000:2000::28<br />
Virtual control room<br />
(GRNET)<br />
2001:648:2320:7:250:56FF:FE9B:4806<br />
VCR<br />
Ipv6 connection<br />
Seismic<br />
Network<br />
Unversity <strong>of</strong><br />
Genoa through<br />
GARR<br />
Ipv4 connection<br />
Figure 23 Test bed Setup<br />
Figure 23 summarizes the basic components <strong>of</strong> the DORII middleware <strong>and</strong> how they interact in this<br />
application. The web portal in the VCR acts as the user interface in order to perform all the<br />
operations. The IE is the gateway to the instrument that in this case is a seismic sensors network.<br />
The IE is in charge <strong>of</strong> the monitoring <strong>and</strong> local storage <strong>of</strong> seismograms, in case that a potential<br />
seismic event is detected.<br />
The user launches the following computation procedure: the files are moved to a SE in order to be<br />
accessed from everywhere in the grid domain. The executable is passed to a CE, which then stores<br />
the results back to the SE. The results are the input seismic data in binary format provided with the<br />
P-wave onset time reference. Moreover, the details <strong>of</strong> the event <strong>and</strong> the P wave are shown to the<br />
user through the VCR. It is clear that the workflow is made up <strong>of</strong> two different parts: interaction<br />
with instrument <strong>and</strong> computation.<br />
Furthermore, Figure 24 shows a snapshot concerning the monitoring <strong>of</strong> the IPv6 traffic between<br />
EUCENTRE <strong>and</strong> GRNET during a testing period.<br />
Finally, Figure 25 highlights a traffic trace between the VCR <strong>and</strong> the IE captured by using<br />
TCPDUMP.<br />
DSA1.5 PUBLIC 24
Figure 24 IPv6 <strong>Traffic</strong> Monitoring<br />
Figure 25 IPv6 <strong>Traffic</strong> traces between the VCR <strong>and</strong> the IE<br />
DSA1.5 PUBLIC 25
6. Conclusions<br />
This document has updated the previous deliverable DSA1.5 with new results <strong>of</strong> experimental tests<br />
concerning two <strong>of</strong> the deployed DORII applications (EEWS <strong>and</strong> OPATM-BFM). It has also<br />
provided a description <strong>of</strong> the IPv6 test bed setup, <strong>and</strong> reported some data on the evaluation<br />
conducted on it.<br />
The results confirm that currently the network is not likely to be a bottleneck for the DORII<br />
applications under examination, <strong>and</strong> that IPv6 connectivity can be enabled with relatively limited<br />
effort, <strong>and</strong> extended globally upon completion <strong>of</strong> a full adaptation <strong>of</strong> gLite <strong>and</strong> security<br />
components.<br />
DSA1.5 PUBLIC 26
7. List <strong>of</strong> Acronyms<br />
AAI<br />
AN<br />
API<br />
ATM<br />
BoD<br />
CE<br />
CERT<br />
CTD<br />
DAC<br />
DS<br />
DSCP<br />
DVB<br />
DVB-RCS<br />
DWDM<br />
EEWS<br />
ER<br />
ETSI<br />
FTP<br />
GDAC<br />
GGF<br />
GHPN<br />
GMPLS<br />
GS<br />
GUI<br />
IE<br />
IETF<br />
INGV<br />
ISP<br />
LAN<br />
LBE<br />
LSP<br />
MAN<br />
MC&A<br />
MDM<br />
MPLS<br />
NCC<br />
NCSS<br />
NOAA<br />
NREN<br />
OGF<br />
OGSA<br />
PDH<br />
PERT<br />
Authorization <strong>and</strong> Authentication Infrastructure<br />
Access Network<br />
Application Programming Interface<br />
Asynchronous Transfer Mode<br />
B<strong>and</strong>width on Dem<strong>and</strong><br />
Computing Element<br />
Computer Emergency Response Team<br />
Conductivity, Temperature, <strong>and</strong> Depth<br />
Data Assembly Centres<br />
Differentiated Services<br />
Differentiated Services Code Point<br />
Digital Video Broadcasting<br />
Digital Video Broadcasting with Return Channel Satellite<br />
Dense Wavelength Division Multiplexing<br />
Earthquake Early Warning System<br />
Edge Router<br />
European Telecommunications St<strong>and</strong>ards Institute<br />
File Transfer Protocol<br />
Global Data Assembly Centre<br />
Global Grid Forum<br />
Grid High Performance Networking<br />
Generalized-MPLS<br />
Guaranteed Service<br />
Graphical User Interface<br />
Instrument Element<br />
Internet Engineering Task Force<br />
Istituto Nazionale Ge<strong>of</strong>isica e Vulcanologia (Bologna, Italia)<br />
Internet Service Provider<br />
Local Area Network<br />
Less than Best Effort<br />
Label Switched Path<br />
Metropolitan Area Network<br />
Measurement, Control & Automation<br />
Multi Domain Monitoring<br />
Multiprotocol Label Switching<br />
Network Control Centre<br />
Network Centric Seismic Simulation<br />
National Oceanic Atmospheric Agency<br />
National Research Network<br />
Open Grid Forum<br />
Open Grid Service Architecture<br />
Plesiochronous Digital Hierarchy<br />
Performance Enhancement <strong>and</strong> Response Team<br />
DSA1.5 PUBLIC 27
PHB<br />
PIP<br />
PoP<br />
PTS<br />
QoS<br />
RFC<br />
RIPE<br />
RIS<br />
RSVP-TE<br />
SAXS<br />
SE<br />
SLA<br />
SMIWR<br />
SOA<br />
SON<br />
SYRMEP<br />
TCP<br />
UDP<br />
VBR<br />
VCR<br />
VLAN<br />
VO<br />
VPN<br />
WAN<br />
Wi-Fi<br />
WN<br />
WOCE<br />
WSN<br />
XRD<br />
Per-Hop Behaviour<br />
Premium IP<br />
Point <strong>of</strong> Presence<br />
PERT Ticket System<br />
Quality <strong>of</strong> Service<br />
Request For Comments<br />
Réseaux IP Europeens<br />
Remote Instrumentation Service<br />
Resource resSerVation Protocol with Tunnel Extensions<br />
Small Angle X-ray Scattering<br />
Storage Element<br />
Service Level Agreement<br />
Simulation <strong>and</strong> Monitoring System for Inl<strong>and</strong> Waters <strong>and</strong> Reservoirs<br />
Service Oriented Architecture<br />
Service Oriented Network<br />
SYnchrotron Radiation for Medical Physics<br />
Transmission Control Protocol<br />
User Datagram Protocol<br />
Variable Bit Rate<br />
Virtual Control Room<br />
Virtual LAN<br />
Virtual Organization<br />
Virtual Private Network<br />
Wide Area Network<br />
Wireless Fidelity<br />
Working Node<br />
World Ocean circulation Experiment<br />
Wireless Sensor Network<br />
X-Ray Diffraction<br />
DSA1.5 PUBLIC 28
References<br />
[1] EU DORII Project - Deliverable DSA1.1, http://www.<strong>dorii</strong>.eu<br />
[2] EU DORII Project - Deliverable DSA1.2, http://www.<strong>dorii</strong>.eu<br />
[3] EU DORII Project - Deliverable DSA1.3, http://www.<strong>dorii</strong>.eu<br />
[4] EU DORII Project - Deliverable DNA3.2 – http://www.<strong>dorii</strong>.eu<br />
[5] EU DORII Project - Deliverable DSA1.5 – http://www.<strong>dorii</strong>.eu<br />
[6] IETF RFC 2330, “Framework for IP Performance Metrics”, http://www.ietf.org/rfc/rfc2330.txt<br />
[7] SmokePing Home Page, http://oss.oetiker.ch/smokeping/<br />
[8] RRDTool, http://oss.oetiker.ch/rrdtool/<br />
[9] Pathload Home Page, http://www.cc.gatech.edu/fac/Constantinos.Dovrolis/bw-est/pathload.html<br />
[10] IETF RFC 1157, “SNMP” http://www.ietf.org/rfc/rfc1157.txt<br />
[11] Nagios Home Page, http://www.nagios.org<br />
[12] NagiosGrapher, http://sourceforge.net/projects/nagiosgrapher/<br />
[13] GARR, The Italian Academic & Research Network, http://www.garr.it/garr-b-home-engl.shtml<br />
[14] RedIRIS, The national research <strong>and</strong> education network (NREN) for Spain, http://www.rediris.es/index.en.html<br />
[15] PIONIER: Polish Optical Internet - Advanced Applications, Services <strong>and</strong> Technologies for Information Society,<br />
http://www.pionier.gov.pl/project/program.htm<br />
[16] DFN, The national research <strong>and</strong> education network (NREN) for Germany,<br />
http://www.dfn.de/index.php?id=74989&L=2<br />
[17] GRNET, The national research <strong>and</strong> education network (NREN) for Greece,<br />
http://www.grnet.gr/default.asp?pid=1&la=2<br />
[18] GÉANT2, The seventh generation <strong>of</strong> pan-European research <strong>and</strong> education network, http://www.geant2.net/<br />
[19] IETF RFC 3493, “Basic Socket Interface Extensions for IPv6”, http://www.ietf.org/rfc/rfc3493.txt<br />
[20] IETF RFC 3542, “Advanced Sockets API for IPv6”, http://www.ietf.org/rfc/rfc3542.txt<br />
[21] IETF RFC 2732, “Format for Literal IPv6 Addresses in URL's”, http://www.ietf.org/rfc/rfc2732.txt<br />
DSA1.5 PUBLIC 29
Contact Information<br />
All authors affiliation:<br />
Poznań Supercomputing <strong>and</strong> Networking Center<br />
ul. Noskowskiego 10<br />
61-704 Poznań, Pol<strong>and</strong><br />
URL: http://www.<strong>dorii</strong>.eu/<br />
Franco Davoli<br />
Davide Adami<br />
Anastasios Zafeiropoulos<br />
Stefano Vignola<br />
franco.davoli@cnit.it<br />
davide.adami@cnit.it<br />
tzafeir@admin.grnet.gr<br />
stefano.vignola@cnit.it<br />
DSA1.5 PUBLIC 30