Transcript Measurement tools and the adoption of Abilene Observatory

Measurement tools
and the adoption of
Abilene Observatory
APAN Cairns meeting, July 2004
Jin Tanaka & Takatoshi Ikeda
APAN-JP NOC
Outline
 Background
 Measurement Tools in APAN-JP
 Selection and Provision of Measurement
data
 Deployment plan of Observatory
Background
 Various measurement tools are installed independently by
research projects in APAN-JP.
Operators can’t grasp the whole measurement tools.
Some demonstrations are being done without any notifications.
 Multiple measurement tools sometimes deal with the same
data simultaneously.
We can see the cases that the same data are measured with multiple
machines.
Many researchers have paid attention to throughput data and routing
information as the research material.
 Operators are investigating the network condition in spots.
Operators are required to provide the researchers not only with connectivity
but with more advanced services as needed.
Operators are required to collaborate with the researchers, just before the
events or the demonstrations.
Members of APAN-JP NOC
<Categories>
Network Researchers
<Job Functions>
Experiments &
Advanced Services
Network Engineers
Basic Services
including Performance
24 x 7 Operators
Trouble Tickets
Measurement Tools in APAN-JP
Throughput -1-
Machine name
Installed
tools
Function
Project
Administra
tor
Note
Perf3
Iperf
Bandwidth
measurement
Internet End-toend Performance
Monitoring Bandwidth to the
World (IEPMBW) project
KDDI Labs
http://www.slac.stanfo
rd.edu/comp/net/band
widthtests/hercules/html/sc
atplots/node1.jp.apan.
net.html
Perf4&7
Iperf
Bandwidth
measurement
Operation tool of
NOC
KDDI Labs
-
perf5&
perf6&
genkai-perf1&
genkai-perf2
Web100*¹
FastTCP
Iperf
BWCTL
Original tools
Bandwidth
measurement
Highperformance
network for
eVLBI
M.Hirabaru
(NICT)
http://mrtg.koganei.itr
c.net/
*¹
http://www.web100.org/
Measurement Tools in APAN-JP
Throughput -2-
Machine name
Installed
tools
Function
Project
Administra
tor
fs01&fs02
FastTCP
Bandwidth
measurement
FastTCP test
across TransPAC
R.Jayaram
S.Low
(caltech)
?
GNET-1*²
Bandwidth
measurement
Bandwidth
challenge
O.Tatebe
(AIST)
bwctl
Iperf
BWCTL
Scheduling and
resource
allocation
wrapper for Iperf
piPEs
Y.Kitamura
(NICT)
*² http://www.gtrc.aist.go.jp/gnet/
Note
-
https://datafarm.apgri
d.org/software/
-
Measurement Tools in APAN-JP
Routing
Machine name
Installed
tools
Function
Project
Administra
tor
Note
rv
mrtd
Routing
information (BGP)
APII BGP collector
for APAN
Measurement WG
M.Hirabaru
(NICT)
http://mrtg.koganei
.itrc.net/
maru1 &
maru2
Zebra *
Tcpdump
Original tools
Routing
information (BGP)
Analysis of the
relation between
traffic and routing
M.Maruyama
(U-Tokyo)
http://maru2.jp.apa
n.net
ope1
net-snmp
Tcpdump
Original tools
Routing
information
(OSPF,IS-IS)
Research for
IGP stability
Z. Sheu
(NICT)
http://ope1.jp.apan
.net/cgi-bin/lsastat?net=apan&area
=backbone
ns2
Original tools
Routing
information(BGP)
Operation tool of
NOC
KDDI Labs
http://ns2.jp.apan.
net/nr/
http://ns2.jp.apan.
net/bgproute/
*
http://www.zebra.org/
Measurement Tools in APAN-JP
Latency
Machine
name
Installed
tools
Function
Project
Administra
tor
Note
?
GNET-1
One-way Delay
measurement
Bandwidth
challenge
O.Tatebe
(AIST)
https://datafarm.a
pgrid.org/software/
pe2(koganei.
wide.ad.jp)
qpe
OWAMP
One-way Delay
measurement
Research for OneWay Delay
Z. Sheu
(NICT)
http://pe2.koganei.
wide.ad.jp/cgibin/owd-stat
ns2
Smoke
ping*⁴
Round Trip Time
measurement
Operation tool of
NOC
KDDI Labs
http://www.jp.apan
.net/measurement/
rtt/
*⁴ http://people.ee.ethz.ch/~oetiker/webtools/smokeping/
Measurement Tools in APAN-JP
Router
Machine
name
ns
Installed
tools
CVS
Original tools
Function
Config backup
and history
Project
Operation tool of
NOC
Administra
tor
NOC Engineer
Note
-
Measurement Tools in APAN-JP
Netflow
Machine name
nms3
Installed
tools
Flow tools
Function
Netflow
Project
Observatory
Administra
tor
KDDI Labs &
NOC Engineer
Note
-
Measurement Tools in APAN-JP
Syslog
Machine
name
ns
Installed
tools
syslogd
Original tools
Function
Reporting
router
information
Project
Operation tool of
NOC
Administra
tor
NOC Engineer
Note
-
Measurement Tools in APAN-JP
Usage
Machine name
Installed
tools
Function
Project
Administra
tor
Note
ns
MRTG
RRDTool
Traffic
measurement
Operation tool of
NOC
NOC Engineer
http://www.jp.apan.
net/NOC/tm/
mrtg
Cricket
RRDTool
Original tools
Traffic
measurement
Operation tool of
NOC
NOC Engineer
http://mrtg.jp.apan.ne
t/cricket/routerinterfaces/
mrtg.koganei.
itrc.net
MRTG
Original tools
Traffic
measurement
M.Hirabaru
(NICT)
http://mrtg.koganei.itr
c.net/
Selection and Provision of
Measurement Data
 Sort out data required for researchers
 Throughput
 Iperf, BWCTL
 Latency
 OWAMP,RRD
 Netflow
 cflow,flowtools
Evaluation
＆
Selection
 GNET-1
 IGP Routing
information
 Tcpdump
… etc.
to be provided by the
Observatory system that
NOC members will
coordinate.
Advanced backbone
services
to be adopted into the
Observatory system as
“new” common tools.
to be left as
the private tools.
Collocated systems
assisted by NOC
Engineers
Selection and Provision of Measurement Data
Potential “New” Common Tool
 ~Example~
 GNET-1
Developed by AIST (The National Institute of Advanced
Industrial Science and Technology, Japan)
GNET-1 is a gigabit network testbed, which consists of four
Gigabit Ethernet interfaces, four fast and large amount of
SRAMs and a large-scale FPGA connected to the them.
Function




Wide area network emulation
Network instrumentation
Traffic shaping
Traffic generation
Powerful tool for developing
network-aware grid software
like GridMPI and Gfarm
Selection and Provision of Measurement Data
TransPAC/APAN Observatory Systems
 We are planning to provide the services of two
categories:
 Advanced backbone data
 Backbone data collected by NOC using equipment located in
TransPAC/APAN nodes. Data item and data format will become
the same as those of Abilene observatory.
Just started by the Joint TransPAC and APAN-JP NOC members.
 Collocation Systems assisted by NOC engineers
 Data collected by respective research projects using their own
equipments collocated in XP.
 Accounts should be provided for NOC operators and engineers.
 Permission of exporting data should be authorized by the projects.
Selection and Provision of Measurement Data
Backbone Data Collection
 Necessity for backbone measurement
 Advanced Backbone measurement is required for improving NOC
services.
NOC operators can’t coordinate the E2E measurement.
 Advanced data enables to allocate the network resources effectively.
Congestion of measuring activities will be relieved.
 It helps operators to find “where is a bottle-neck?”
 It activates network researches with advanced backbone data.
Observatory
Backbone
Measurement
E2E Measurement
Selection and Provision of Measurement Data
Proposal of TransPAC/APAN
Observatory
 Measurement should be done in the best conditions
 Observatory systems should be installed in the best environments of
TransPAC/APAN network.
i.e. To be set as close to the router interfaces as possible, using the best
interface cards of Jumbo Frame.
 NOC should maintain the systems and export data to authorized
researchers.
 A good common platform for performance measurement
should be installed at each XP.
 Common measurement method should be developed
 Common tools are installed, providing the common data.
 Data should be formatted in the global standards.
i.e.
GGF, piPEs, IPPM(IETF), IRPM …etc.
Selection and Provision of Measurement Data
Formation made by Observatory
Observatory system will grow up R&E networks!
Software,
Papers, etc.
NOC Basic Service
NOC Advanced Service
Observatory
Researchers
•Developed by NOC researchers &
maintained by NOC engineers/operators
•Common Tools with High Priority
•Output data in standard format
can get measurement data of the
global networks and work together
with foreign researchers.
Deployment plan of Observatory
Observatory system could be deployed
across the TransPAC circuits.
This deployment is to :
 get data over the long latency link
 cooperate with the Abilene Observatory
Deployment plan of Observatory
 Location
 Japan
 Tokyo XP
 U.S.
 To be decided by Indiana University folks
Chicago/Indianapolis
TokyoXP
Los Angeles
Average RTT 190ms
Deployment plan of Observatory

Machines

There are ４ machines in each side for observatory. (NMS1-4)
The configuration is the same as that of Abilene Observatory :




optimized for observatory based on their experience
made easy to cooperate with the Abilene Observatory
managed by NOC
Hardware Specification







Chassis : SuperMicro SuperServer 6013P
CPU : Dual 3.06GHz Intel Xeon
Memory : 2 x 1GB DDR266
Disk : 2 x SCSI 73GB (RAID1)(NMS2~4）
3 x SCSI 73GB (RAID5)(NMS1）
NIC : Dual 10/100/1000 on-board
Intel 1000base-SX （NMS1)
Deployment plan of Observatory
 Roles of NMS
host
OS
Measurement
tool
Network
NMS1
Linux 2.4.26
web100 2.3.8
Throughput
(regularly)
iperf
GbE to router &
switch
MTU 9000
NMS2
Linux 2.4.26
web100 2.3.8
Throughput
(on-demand)
iperf
GbE to switch
MTU 9000
NMS3
FreeBSD 5.2.1-RELp8
Netflow
Usage
Flow-tools
Net-snmp +RRD
GbE to switch
MTU 9000
NMS4
FreeBSD 5.2.1-RELp8
Latecy
OWAMP
GbE to switch
MTU 9000


Routing, Router, Syslog and Usage data will be evaluated for the deployment.
NICs on all servers are for Gigabit Ethernet.
Deployment plan of Observatory
 Network Layout (TokyoXP)
Procket PRO/8812
GigE Fibre
NMS1
GigE Metal
NMS2
GigE Metal
NMS3
GigE Metal
GigE Fibre
NMS4
GigE Metal
Deployment plan of Observatory
 Rack Front View
Deployment plan of Observatory
 Rack Rear View
Current Status
 Current Status (TokyoXP)


The machines has installed the OS and connected the network
Ready to collect data

Netflow
 Ready to collect a Netflow data with Juniper router

Problem with the Procket routers:
The Procket routers create the UDP data of MTU size for netflow,
IP packet containing UDP header and IP header exceed the
MTU size. Hence fragmentation is occurred and flow-tools
cannot receive packets correctly.
->will be fixed in the next software version

Throughput and Latency
 Ready to collect throughput and latency data with the peering
machine located in Chicago, managed by John-san.

Need to tune some parameter for good throughput data
(buffer , flow-control )

Our L2 switch didn’t perform wire-rate switching. Some packets
was discarded at input port while testing a high bit rate . we
will investigate in detail .
Roadmap
 under discussion with Indiana University,
but the following schedule is envisaged:
 2004.7
Netflow & piPEs (temporal)
Website was formed:
http://www.jp.apan.net/NOC/Observatory/
 2004.8 start to develop the real system
 2004.10 Connection with Abilene Observatory
 2005.1 Presentation at APAN Bangkok meeting
Deployment into APAN
 We expect that Observatory system will be
widely deployed in APAN.
TransPAC Observatory
Korea

Thailand
Malaysia
Tokyo
North America
USA
 Genkai
China
Hong Kong 


Taiwan

Vietnam

Singapore
Philippines
Indonesia
APAN Observatory
 Australia
References
 Abilene Observatory


http://abilene.internet2.edu/observatory/
http://loadrunner.uits.iu.edu/~neteng/nms/
 Tools







web100 http://www.web100.org/
GNET-1 http://www.gtrc.aist.go.jp/gnet/
zebra http://www.zebra.org/
Smokeping
http://people.ee.ethz.ch/~oetiker/webtools/smokeping/
Flow-tools http://www.splintered.net/sw/flow-tools/
OWAMP http://e2epi.internet2.edu/owamp/
Iperf http://dast.nlanr.net/Projects/Iperf/
Thank you
 Please feel free to contact us if you
have any questions and comments
 Jin Tanaka – [email protected]
 Takatoshi Ikeda – [email protected]