Test before starting service
Download
Report
Transcript Test before starting service
Toward Stable Operation
of Genkai/Hyeonhae/APII link
Yoshinori Kitatsuji
APAN Tokyo XP
KDDI R&D Laboratories, Inc.
Overview
•
Test before starting service
–
•
Staff in Service
–
–
–
•
Routes
Link Utility
Bandwidth availability
Troubleshoot
–
–
•
Checking Availability
Tools
status to public
Future arrangement
–
Resource management
•
Scheduling
–
–
•
Project
Maintenance
Making requirement clear
–
Protocol, Path, Traffic bandwidth, Schedule
Test before starting service
• Understanding availability is much important
– Spec. written in Catalogue doesn't’ match with actual performance
sometime.
– Combination of gears usually perform like the greatest common measure.
– It’s comfortable for both users and operators to use networks if the
limitation is know well.
• It helps scheduling upgrade network and enhancing the performance.
• Performance known at the moment
– Meinohama-Busan
• 930Mbit/s (IP packet, 1000 byte length)
– Meinohama-Tokyo
• 250Mbit/s (UDP/TCP), because shaper (271Mbit/s) applied
• 430Mbit/s (TCP over a loop Tokyo-Tenjin-Tokyo)
– Should be known how much improved
• between QGPOP and Seoul
• between Seoul and Tokyo
Genkai Network Configuration at L3
Busan XP
Tokyo XP (KDDI Otemachi)
AS 9270
ATM/OC12
tpr3
Cisco GSR
169
430Mbit/s
161
tpr2
106
194
AS 7660
203.181.249.160/29
AS 7660
203.181.249.104/30
203.181.249.184/29
203.181.249.168/29
203.181.248.192/30
162
185
APIIJuniper
ATM/OC3
250Mbit/s
105
170
GbE
177
GbE
fidc203.181.249.176/29
juniper
133.69.164.0/28
178
13
GbE
AS 2523
Kyushu U
AS 2523
193
6 hakozakicisco2
GbE
Genkai XP (Meinohama)
AS 2907
GbE
133.69.152.0/26
15
hakozakijuniper
Super
SINET
Presented in NOC meeting, APAN Fukuoka Meeting
Configuration of Loopback test
OC12
Orenge: ATM Switch
Blue: Ethernet Switch
CRL
OC12
APII Juniper
JGN
ATM network
PC
CRL
OC12
APAN
Tokyo
XP
OC12
CRL
OC12
APAN
Tokyo
XP
GigabitEthernet
OC12
OC12
PC
Staff prepared in service
• Making acceptable usage policy clear
– Simple and Opened AUP helps for operators to manage trouble.
• Monitoring in APAN Tokyo XP
– Traffic, Number of routes
• MRTG, RRDTool, OCXmon/DAG
– Reachability
• Ping and report for major targets every 5minutes
– Performance
• SLAC’s performance monitor using iperf
• Opening operational information to public
– Assignment information
• IP address, ATM PVC, VLAN ID, Port utility of equips
– Monitoring result
MRTG
•
Multi Router Traffic Grapher
–
–
–
–
•
Advantages
–
–
–
–
•
Most common tool for operation
Plots 2 data source against time
Use SNMP to collect counters of interfaces
5min average and 2 data sources in a figure
Ability to create scripts to feed data into MRTG
Mature code because of very wide deployment in large networks
Any variable can be graphed
Easy to understand configuration
Disadvantages
–
–
–
Can’t create custom graph periods
Only keeps 5 minute data for 2.5 days, after witch it is aggregated to longer period.
Too hard to hack to modify higher late than 5 minutes
RRDTool
•
Round Robin Database Tool
–
–
–
•
Advantages
–
–
–
•
Successor of MRTG
Ability to customize graphs across user-defined interval
Constant file size independs on the number of feeding time because of binary file format
25-30 percent faster than MRTG
Ability to have multiple data sources in a single graph.
Ability to generate graph from multiple data files.
Disadvantages
–
–
–
No collector
Doesn’t create graph and web-pages without additional scripts
Hard to merge multiple files into one.
CoralREEF/DAG
•
TCPDUMP for ATM/POS
–
–
Packet capturing system
POS OC48 is under development
•
–
CoralREEF
•
•
–
Successor of OC3mon developed by MCI & NLANR
run with Fore PCI200 or Point ATM card
DAG
•
•
•
Unknown how much perform actually
developed by Wikato University in New Zealand
run with DAG card developed by Wikato University
Router
Advantages
–
–
Header or whole packets are captured
Ability to convert original format to TCPDUMP (pcap
library)
ATM Switch
•
Disadvantages
–
–
–
Run on only FreeBSD or Linux
Splitter divide the laser and power degrades to half
Need multiple PCs to capture a high speed link.
•
–
May need GPS clock than NTP to get high accuracy
Need interface to generating graphs continuously like
MRTG
Traffic since …
Hyeonhea/Genkai Link
Here operation started
TransPAC link to Seattle
Here Genkailink operation started
SLAC performance monitor
Here operation started
• TCP performance monitor
• Disadvantage
─ Generate TCP traffic for 1-2 minutes.
─ Generated traffic effects regular traffic.
─ Measure RTT and bandwidth of
─ Takes some time (2-3minutes) to measure.
iperf, BBCPMEM, BBFTP
APAN Tokyo XP
─ http://perf3-fe.jp.apan.net/~cottrell/html/slac_wan_bw_tests.html
• Advantage
− requires only ssh and iperf
− multiple targets are measured round robin every
two hours automatically
Stanford Univ
QGPOP
AIST
Open Operation
• Opening assignment policy and operation status
– Reducing query
– Make troubleshoot easier
• Sharing account between
neighbor networks
– Consider to cause the lack of security,
but fruitful.
– Ability to understand neighbor networks deeply
– Account holder do troubleshoot by themselves
• Opening status of trouble
– Users and neighbor networks feel relieved.
– May have help from outside
Troubleshoot
• Tool
– ping/traceroute for unicast
– mtrace, multicast beacon and debug output for multicast
– MRTG or more accurate graph
• Sharing information
– Maintenance report
• schedule, status, and after maintenance
• Traceable status board
– Make any information opened as much as possible
• Neighbor may do troubleshoot
Multicast Beacon
• Every beacon sends and
receives periodical packets to
some group and all of them
make reports the statistics to
MB server.
– Developed by NLANR
• http://dast.nlanr.net/Projects/B
eacon/
– Popular in AccessGrid
• MB server in XP may helps to
debug the multicast trouble for
especially seeking where
problem lie.
Future operation
• Resource Scheduling will become important
– Now we can generate 100s Mbit/s speed traffic between Korea and
Europe through Japan and USA
– More important to schedule events than sharing resources over events
– Multiple Traffic easily break each other.
• Will take more time to support the Events
– Demonstration will requires the high performance
• Ex. AIST tried to record high speed TCP transmission test with 2 TransPAC
links in parallel in SC2002.
– Need to Collect detail of requirement
– Regular traffic increase gradually
• Installing Tools and Devices
– XPs are required to equips tools and devices to support all kind of services
run by projects