internet2 performance tools fall 2011 iips conference - … · 2017-02-20 · 2 10/26/11 ....
TRANSCRIPT
Carla Hunt Manager Knowledge and Information Systems, MCNC [email protected]
Tom Throckmorton Sr. Network Management and Information Systems Architect, MCNC [email protected]
Internet2 Performance Tools Fall 2011 IIPS Conference
Overview
Introduction and Motivation for Performance Tools (Carla)
On Demand Throughput Measurements (Carla)
One Way Latency Measurements (Tom)
Internet2 perfSONAR Performance Toolkit (Tom)
2 10/26/11
Introduction
About Us
MCNC is the company and NCREN is the network, our !agship product. • MCNC is an independent, non-pro"t organization that employs
advanced networking technologies and systems to help various sectors of Community Anchor Institutions (CAIs) in North Carolina communicate with their constituents more effectively and meet their speci"c organization's mission, vision and goals.
Knowledge and Information Systems at MCNC • The mission of the Knowledge and Information Systems team is to
establish an overall strategy for knowledge capture and dissemination and to develop and provision the MCNC knowledge capture and dissemination infrastructure.
4 10/26/11
About Internet2
One of two national research networks to which NCREN connects. We send research traffic and subscribe to a commercial peering service.
Over the last ten years Internet2 has also led the development of a measurement framework and collection of measurement tools
• perfSONAR-PS • pS-Performance Toolkit
5 10/26/11
About Internet2 and MCNC
Relationship extends beyond the network • Carla, co-chair of Internet2 Performance WG • Tom, contributor to ps-PS development
MCNC and Internet2 offered a 1.5 day network performance workshop earlier this month for system and network administrators to learn more about these tools
• 100% of participants surveyed thought workshop a valuable use of time
• 1.5 days into 1hr 15min
6 10/26/11
Motivation
Are you responsible for a college network?
Are you or have you experienced network performance problems?
Do you or have you ever had one or more network users who say the network is slow or broken?
Would you like to be able to monitor and track short-term and long-term network statistics to identify trends and for capacity planning?
7 10/26/11
About this session
There is no single tool or approach to diagnose network performance issues
We’re here to talk about tools, strategies and recommendations that will help
Patience and diligence will get you to most goals
8 10/26/11
Internet2 End to End Performance Initiative
Traditional network performance data collection centralized – think Cacti or MRTG
Approach network performance in the context of multiple administrative domains
9 10/26/11
What do we mean by multiple domains?
10 10/26/11
User typically somewhere on your campus but the resource they are trying to access could be across campus or on a part of the network you don’t manage
Internet2 pS-Performance Toolkit
What is it? • A collection of open source tools that can be
easily deployed on your network • A framework for sharing and discovery of
network performance data • A resource for users and admins to perform on
demand and scheduled measurement tests
We will focus on tools that enable throughput and latency measurement but there are other tools in the collection
11 10/26/11
Internet2 pS-Performance Toolkit (cont.)
Low cost and low effort to deploy
Measurement endpoints are discoverable
These tools have been used to solve regional and international network performance issues, as well as locally within NCREN!
If you have had a network health assessment, we have used some of them on your campus network
12 10/26/11
Myths and Pitfalls
• “My LAN performance is great, WAN is probably the same” – TCP recovers from loss/congestion quickly on the LAN (low RTT) – TCP will cut speed in half for every loss/discard on the WAN – will take a
long time to recover for a large RTT/ – Small levels of loss on the LAN (ex. 1/1000 packets) will go unnoticed,
will be very noticeable on the WAN.
• “Ping is not showing loss/latency differences” – ICMP May be blocked/ignored by some sites – Routers process ICMP differently than other packets (e.g. may show
phantom delay) – ICMP may hide some (not all) loss. – Will not show asymmetric routing delays (e.g. taking a different path on
send vs receive)
• Our goal is to dispel these and others by educating the proper way to verify a network – we have lots of tools at our disposal but using these in the appropriate order is necessary too
13 10/26/11
When problems exist, it’s the networks fault! • Easy to blame a resource, but where else could a problem
be? - Host (Disk, CPU, Kernel, NIC Drivers) - Network Interface Cards - Routers/Switches, Routing and Con"guration - Physical Infrastructure - Protocols
The network is viewed as a single resource in many cases • Reality – complex series of components • Multiple vendors/technologies • Multiple con"guration options • Crossing administrative domains
14 – 10/26/11, © 2011 Internet2
Underlying Assumption
Network View (Layman’s Terms)
15 – 10/26/11, © 2011 Internet2
Bob’s Host “The Internets”
Carol’s Host
Network View (Actual)
16 – 10/26/11, © 2011 Internet2
Sw
itch
1
Switch 2 Switch 3
R1 R3
R4
R2 R7
R6 R9
R8 R5
Switch 4
Finding a solution to network performance problems can be broken into two distinct steps: • Use of Diagnostic Tools to locate problems
- Tools that actively measure performance (e.g. Latency, Available Bandwidth) - Tools that passively observe performance (e.g. error counters)
• Regular Monitoring to establish performance baselines and alert when expectation drops. - Using diagnostic tools in a structured manner - Visualizations and alarms to analyze the collected data
Incorporation of either of these techniques must be: • ubiquitous, e.g. the solution works best when it is available
everywhere • seamless (e.g. federated) in presenting information from different
resources and domains
17 – 10/26/11, © 2011 Internet2
Methodology
What are the "rst steps to address problems related to network performance? • Try a Tool
What tools are out there • Numerous • Different metrics (measurements) available • How to interpret the results?
18 – 10/26/11, © 2011 Internet2
Addressing a Performance Discrepancy
Ping
Traceroute
Iperf
Tcpdump
Tcptrace
BWCTL
NDT
OWAMP
AMP
Advisor
Thrulay
Web100
MonaLisa
pathchar
NPAD
Pathdiag
Surveyor
19 – 10/26/11, © 2011 Internet2
Tools, Tools, Tools
• Ethereal • CoralReef • MRTG • Skitter • C!owd • Cricket • Net100 • Pathload • Pathchrip • MRTG • Cacti • Smokeping • PingER • FDT • perfSONAR • Nagios • Ganglia • Thurlay • Etc. etc. etc.
Focus on 3 Types of tools (for now) • Basic Diagnostics
- Ping, Traceroute • Advanced User Tools
- NDT • Network Admin Focused
- OWAMP, BWCTL
What about the others? • Try them out, learn how they work. • Most tools are designed to solve a speci"c problem and they may
add value to your organization
Integration of multiple solutions • Measurement frameworks integrate use of tools (operation,
collecting results) along with analysis and presentation • perfSONAR
20 – 10/26/11, © 2011 Internet2
Highlighting some Interesting Tools
Ping • Round Trip (e.g. source to destination, and back) • Con"rms that remote host is ‘up’ • Some network operators block these packets - Play w/ command options to see if that will change anything
Traceroute • Identi"es the routers along the path • Same blocking problem as above • Routers treat ICMP packets with lower priority - See presentation from prior JTs: - http://www.internet2.edu/presentations/jt2009jul/20090722-
litvanyi.pdf
21 – 10/26/11, © 2011 Internet2
Basic Diagnostic Tools
NDT = Network Diagnostics Tool • Perhaps some of you in this room have used an NDT server
Measure performance to users desktop
Identify real problems for real users • Network infrastructure could be the problem • Host tuning issues could be the problem
Simple to use and somewhat simple to understand • Presentation in a method almost all users can access: web browser
Tool is designed to be useful for users and network administrators • Variables for many aspects of host, protocol, and network performance
22 – 10/26/11, © 2011 Internet2
Advanced User Tool - NDT
BWCTL – Bandwidth Control • Allows single person operation over wide area testing environment • Inject traffic on the network between two measurement nodes and
measure throughput • Runs NLANR ‘iperf’ program
- Support for Thrulay, nuttcp
OWAMP – One way Delay Measurement • Advanced ‘ping’ command
- One way vs round trip
• Inject traffic on the network between two measurement nodes and measure latency or delay
• Allows single person operation over wide area testing environment
23 – 10/26/11, © 2011 Internet2
Network Administration Tools
pS Performance Toolkit ISO • All tools, pre-installed and con!gured • More info: http://psps.perfsonar.net/toolkit
• Source Packages (Client and Server) • http://software.internet2.edu/sources/ • Typical ‘con"gure/make/make install’
RPM Installation (CentOS 5.5 Supported): • Install our RPM package to enable the Internet2 Repository • See instructions here: http://software.internet2.edu/ • Support for YUM and APT-RPM
Others Notes: • Other RPM based distros (Fedora/RHEL) may work with packaged
RPMs … YMMV • To install on Debian, consider source. Alien conversions of RPMs
may be problematic
24 – 10/26/11, © 2011 Internet2
Software Availability
On Demand Throughput Measurement Using the Network Diagnostic Tool (NDT)
Advanced User Tool - NDT
Best on demand throughput test we know about • http://mitas.csail.mit.edu/papers/
Bauer_Clark_Lehr_Broadband_Speed_Measurements.pdf
We aren’t the only ones that use it • FCC – National Broadband Speed Test -
http://www.broadband.gov/qualitytest/ • Measurement Lab (M-Lab) - http://measurementlab.net/ • Verizon - http://my.verizon.com/micro/speedtest/broadband/
Limitations • Language returned in heuristics may be out of date
26 10/26/11
M-Lab (Commodity Networking) • http://www.measurementlab.net/run-ndt
Internet2 (R&E Networking) • http://ndt.atla.net.internet2.edu:7123/ • To not overwhelm the server, also try replacing ‘atla’ with:
- chic - hous - kans - losa - newy - salt - seat - wash
NCREN NDT Service • https://ndt.ncren.net
27 – 10/26/11, © 2011 Internet2
Hands on Tes6ng of NDT
Web-based JAVA applet allows testing from any browser • One Click testing
• Option to dig deep into available results • Send report of results to network administrators
Command-line client allows testing from remote login shell • Same options available
• Client software can be build independent of server software
28 – 10/26/11, © 2011 Internet2
NDT User Interface
NDT Results
29 – 10/26/11, © 2011 Internet2
Measure performance to users desktop • Lots of tools to measure performance to a nearby server • Also ‘plugable’ hardware to measure everything up to the network
cable • Want something to accurately show what the user is seeing
Develop “single shot” diagnostic tool that doesn’t use historical data
Combine numerous Web100 variables to analyze connection
Develop network signatures for ‘typical’ network problems • Based on heuristics and experience • Lots of problems have a smoking gun pattern, e.g. duplex mismatch,
bad cable, etc.
30 – 10/26/11, © 2011 Internet2
Motivation for Developing NDT
Simple bi-directional test to gather end to end data • Test from client to server, and the reverse
• Gets the ‘upload’ and ‘download’ directions
Gather multiple data variables from server • Via Web100, also some derived metrics (packet inter arrival times)
Compare measured performance to analytical values • How fast should a connection be given the observations of the
host and network
Translate network values into plain text messages
Geared toward campus area network
31 – 10/26/11, © 2011 Internet2
How It works
Operates on Any client with a Java enabled Web browser • No additional client software needs to be installed • No additional con"guration required
What it can do: • State if Sender, Receiver, or Network is operating properly • Provide accurate application tuning info • Suggest changes to improve performance
What it can’t do • Tell you exactly where in the network the problem is • Tell you how well or poorly “other” servers perform • Tell you how well or poorly “other” clients will perform
32 – 10/26/11, © 2011 Internet2
Web Based Performance Tool
Duplex Mismatch • This is a serious error and nothing will work right. Reported on
main page, on Statistics page, and mismatch: on More Details page
Packet Arrival Order • Inferred value based on TCP operation. Reported on Statistics
page, (with loss statistics) and order: value on More Details page
Packet Loss Rates • Calculated value based on TCP operation. Reported on Statistics
page, (with out-of-order statistics) and loss: value on More Details page
Path Bottleneck Capacity • Measured value based on TCP operation. Reported on main page
33 – 10/26/11, © 2011 Internet2
Finding Results of Interest
What is the slowest link in the end-to-end path? • Monitors packet arrival times using libpcap routine
- Data and ACK packets - Is aware of packet sizes – used to calculate speed
• Use TCP dynamics to create packet pairs • Quantize results into link type bins
- Broad classi"cation, e.g. “FastE” - No fractional or bonded links currently
Example: • Consider the following setup
- 1G network card on Host - 1G LAN - 100M (FastE) Wall Jack
• NDT will report there is a slow link somewhere in the path. It can’t tell you where, but something is limiting the test speed
34 – 10/26/11, © 2011 Internet2
Bottleneck Link Detection
Duplex Mismatch: • Operation between a host and an interface are at different
duplex modes (e.g. one half, one full) • Common in networks where auto negotiation is disabled, or
faulty • Classic example of a “soft failure”, connectivity is present and
speeds are poor
35 – 10/26/11, © 2011 Internet2
Duplex Mismatch Detection
Detect non-congestive loss due to • Faulty NIC/switch interface
• Bad Cat-5 cable • Dirty optical connector
36 – 10/26/11, © 2011 Internet2
Faulty Hardware or Link
Shared network infrastructures will cause periodic congestion episodes • Detect/report when TCP throughput is limited by cross traffic • Detect/report when TCP throughput is limited by own traffic
37 – 10/26/11, © 2011 Internet2
Congestion Detection
Provide basic tuning information
Features: • Basic con"guration "le • FIFO scheduling of tests, support for testing with simultaneous
clients • Simple server discovery protocol and ability to federate (e.g. load
balance) servers • Logging of all test results on the server side
Command line client support
Other Clients can be developed against open Javascript API: • http://www.internet2.edu/performance/ndt/api.html
Posted on Google Code: • http://code.google.com/p/ndt/
38 – 10/26/11, © 2011 Internet2
Additional Functions and Features
Architecture
39 – 10/26/11, © 2011 Internet2
Client Web
Browser
Java
Applet
NDT - Server
Web
Server
Testing
Engine
Child
Test Engine
Spawn child
Well Known NDT Server
Web Page Request
Web page response
Test Request
Control Channel
Specific test channels
There are no settings or options for the Web based java applet. • It allows the user to run a "xed set of tests for a limited time
period
Test engine settings • Turn on admin view (-a option) • If multiple network interfaces exist use –i option to specify
correct interface to monitor (ethx)
Simple Web server (fakewww) • Use –l fn option to create log "le
• Could also use a ‘real’ web server like Apache
40 – 10/26/11, © 2011 Internet2
Recommended Settings
Non-standard kernel required • Web100 patching may be difficult to apply to new kernels
• Hard to keep up with vendor patching • GUI tools can be used to monitor other ports • Consider using pS Performance Toolkit enhancements if this
scares you…
Public servers generate trouble reports from remote users • Respond or ignore emails
Test streams can trigger IDS alarms • Con"gure IDS to ignore NDT server
41 – 10/26/11, © 2011 Internet2
Potential Risks
One Way Latency Measurement Using OWAMP
OWAMP is: • Command line client application • Policy and scheduling daemon • Used to determine one way latencies between hosts.
Implementation of the OWAMP protocol as de"ned by http://www.rfc-editor.org/rfc/rfc4656.txt • Command Protocol to speak between client and server, server and
server • Test protocol
Different attempts to do this in the past: • Surveyor • RIPE
43 – 10/26/11, © 2011 Internet2
OWAMP: What is it
Passive Measurements alone not enough (e.g. SNMP) • Higher polling interval may mask queue depths
• Active probing gives a better picture of real traffic
Round Trip Measurements: • Pros
- Can be done with a single ‘beacon’ (e.g. using ICMP responses) - e.g. latency measurements in the NCREN Community Portal are round
trip
• Cons - Hard to isolate the direction of a problem - Congestion and queuing can be masked in the "nal measurement - ICMP packets are given a low priority, dropped "rst
44 – 10/26/11, © 2011 Internet2
Why One Way Latency?
One Way Measurements: • Direction of a problem is implicit
• Detects asymmetric behavior • See congestion or queuing in one direction "rst (normal
behavior) • Requires ‘2 Ends’ to measure properly
OWAMP Graphs: • Feature two lines for each direction.
• Very sensitive, even to NTP differences • Loss/duplication shows up as an event !ag • Congestive loss as well loss caused by other reasons (equipment
failures)
45 – 10/26/11, © 2011 Internet2
One Way Latency != Round Trip Latency
Different Latencies with pockets of Loss
46 – 10/26/11, © 2011 Internet2
Supports authentication and authorization of the users that will test
Used to con"gure the parameters of a test • Endpoint controlled port numbers • Extremely con"gurable send schedule • Con"gurable packet sizes
Used to start/stop tests
Used to retrieve results • Provisions for dealing with partial session results in the event of a
failure
47 – 10/26/11, © 2011 Internet2
OWAMP Control Protocol
“Lightweight” compared to the control protocol
Uses UDP as the transport protocol, since the protocol needs to be able to measure individual packet delivery times
Supports varying packet sizes
Data needed to calculate experimental errors on the "nal result is in every packet
Packets can be “open”, “authenticated”, or “encrypted”
48 – 10/26/11, © 2011 Internet2
OWAMP Test Protocol
Applications • Daemon (owampd) • Clients (owping, powstream)
Open Source License & Development • Modi"ed BSD (http://www.internet2.edu/membership/ip.html) • Mailing lists for developer communication – come join us!
Protocol Abstraction Library • Will support development of new clients • Add custom ‘hooks’ into the policy (e.g. add authentication
via OpenID or similar)
49 – 10/26/11, © 2011 Internet2
Sample Implementa6on
Meant to operate like traditional “ping”
owping client requests OWD tests from an OWAMP server (owampd)
Client can be ‘sender’ or ‘receiver’ • Both directions are tested unless otherwise speci"ed
Communication can be “open”, “authenticated”, or “encrypted”
Supports the setup of many tests concurrently
Supports the storage of results on the server for later retrieval
50 – 10/26/11, © 2011 Internet2
Func6onality (owping client)
Accepts requests for OWD tests
Responds with accepted/denied
Tests are formally started with a StartSessions message from the client.
Runs tests
Sessions with packets received at the server are buffered for later retrieval
51 – 10/26/11, © 2011 Internet2
Func6onality (owampd server)
OWPING Example
52 – 10/26/11, © 2011 Internet2
OWAMP GUIs – Delay/Loss Plot
53 – 10/26/11, © 2011 Internet2
Similar Latencies, Some Loss
54 – 10/26/11, © 2011 Internet2
OWAMP GUIs -‐ Mesh
55 – 10/26/11, © 2011 Internet2
OWAMP GUIs -‐ JiYer
56 – 10/26/11, © 2011 Internet2
Clock requirement is the strongest • Doesn’t work well in virtualized environments
• Doesn’t work well when machine is doing heavier testing (e.g. BWCTL), results may be suspect
NTP (ntpd) synchronized clock on the local system • Speci"c con"guration requirements as speci"ed in
NTP talk… • Strictly speaking, owamp will work without ntp.
However, your results will be meaningless in many cases
57 – 10/26/11, © 2011 Internet2
OWAMP Requirements
NTP (ntpd) synchronized clock on the local system • Con"gure NTP properly (don’t rely on system defaults!)
• Strictly speaking, owamp will work without NTP. However, your results will be meaningless in many cases
• More information about NTP con"guration here: http://www.internet2.edu/performance/owamp/details.html#NTP
58 – 10/26/11, © 2011 Internet2
General Requirements – Time Source
Plot – Note the “wavy” activity = NTP
59 – 10/26/11, © 2011 Internet2
NCREN NTP Servers
Each rPoP is equipped with a Stratum II time server that allow end users to synchronize the time on individual workstations and other end user devices.
These devices peer with each other and synchronize with our Stratum I servers, to ensure users get the most accurate time.
More information - https://www.mcnc.org/news/network-time-server-upgrade.html
60 10/26/11
“Bare Metal” – virtualization is tricky
Stable System Clock • Temperature controlled environment • No power management of CPU
• Reduction of “background” services – may institute noise
No strict requirements for CPU, Memory, Bus speed • More tasking schedules will require more capable
hardware
61 – 10/26/11, © 2011 Internet2
General Requirements – Hardware
Time: • NTP issues predominate the problems • Determining an accurate timestamp “error” is in many ways
more difficult than getting a “very good” timestamp • Working as an “open” server requires UTC time source (For
prede"ned test peers, other options available)
Firewalls: • Port "lter trade-off - Administrators like pre-de"ned port numbers - Vendor manufactures would probably like to “prioritize” test
traffic - Owampd allows a range of ports to be speci"ed for the
receiver
62 – 10/26/11, © 2011 Internet2
Opera6onal Concerns
Third-Party DoS source • Compromised server may send packets to other locations.
DoS target • Excessive traffic will harm measurement results • Someone might attempt to affect statistics web pages to see how
much impact they can have
Resource consumption • Time slots
• Memory (primary and secondary) • Network bandwidth
63 – 10/26/11, © 2011 Internet2
Policy/Security Considera6ons
Restrict overall bandwidth to something relatively small • Most OWAMP sessions do not require much
Limit “open” tests to ensure they do not interfere with precision of other tests
64 – 10/26/11, © 2011 Internet2
Policy Recommenda6ons
Internet2 pS-Performance Toolkit
Care and Feeding
The pSPT nodes are designed to help two constituencies • Engineers – Regular monitoring to catch problems earlier • Users – Distributed test points to validate their own performance
Work as a system, the following will describe: • The tools • The operations methodology • Maintenance steps
Can integrate with others • Deployed on Backbones • Other regional networks • Other connected members of NCREN
66 – 10/26/11, © 2011 Internet2
Your pSPT Deployment and You
pS Performance Toolkit ISO • All tools, pre-installed and con"gured • Two !avors; LiveCD or NetInstall • LiveCD demo after the presentation
pS Performance Toolkit source • All components are open source and available • packages for RPM-based systems
Consider installing multiple toolkit nodes
Scheduled bandwidth and latency probably won’t go well on the same node
If you install a publicly connected toolkit node and would like to participate in testing with other members of NCREN, we'd suggest you con"gure your node(s) to join the 'NCREN' community. As other test points on NCREN are added, you'll be able to identify them more easily by looking for nodes in that community.
67 – 10/26/11, © 2011 Internet2
Deployment Strategies
Home Page
68 – 10/26/11, © 2011 Internet2
NDT Tests
69 – 10/26/11, © 2011 Internet2
Perfect for the end user to run from their laptop • Convince your campus users to test to this, instead of
‘speedtest.net, or FCC’. Install one on *every* campus • Runs client to server and reverse • As mentioned, Detects common problems
The LiveCD is the quickest way to get NDT deployed on your network
Caveats • Test results on higher bandwidth links can be skewed by poor
client OS tuning; many still shipping with untuned stacks • NDT won’t tell you precisely what’s wrong, but will provide
several clues
70 – 10/26/11, © 2011 Internet2
NDT - quick review
One Way Latency Tests
71 – 10/26/11, © 2011 Internet2
All OWAMP Tests
72 – 10/26/11, © 2011 Internet2
Other tools on the toolkit…
BWCTL • scheduled throughput testing • uses iperf and can perform TCP/UDP tests
Traceroute • Reverse traceroute; regular traceroute testing
NPAD • Detailed path diagnostics
PingER • Scheduled ICMP RTT testing
Cacti • Scheduled SNMP collection
73 10/26/11
Summary
Contact Info
Questions? Need help?
Join the NCREN Network Visualization and Management Discussion Group mailing list: https://list.mcnc.org/mailman/listinfo/net-mgmt-wg
Post to the Technology Forums, https://www.mcnc.org/forums/ncren/performance-monitoring-tools
Internet2 pS-Performance Toolkit • http://psps.perfsonar.net/toolkit
75 10/26/11
76 10/26/11
extra slides (this page intentionally left blank)
OWAMP -‐ Architecture
77 – 10/26/11, © 2011 Internet2
PingER Select Data
78 – 10/26/11, © 2011 Internet2
PingER Graph with spikes of activity
79 – 10/26/11, © 2011 Internet2
Experimental Jitter Graphs
80 – 10/26/11, © 2011 Internet2
Experimental use of OWAMP data
Plots ‘activity’ more than anything. When you see data on the graph its an indication that there is “something” going on • Network congestion • NTP Skew (common) • Host activity (other processes/measurements)
The points do correspond to some amount of “Jitter”, e.g. packet delay variation. • +’s are noise, the larger values may mean something • Red is a large chunk, green are smaller chunks. • Add ‘’alpha_max=0.99” to your string to see a change (e.g. 99th percentile of
Jitter only).
Graphs are still a work in progress, will be ingreated into the toolkit GUIs in the next release
81 – 10/26/11, © 2011 Internet2
Jitter?
Simple Matrix
82 – 10/26/11, © 2011 Internet2
Reverse Traceroute
83 – 10/26/11, © 2011 Internet2
Throughput Measurements
84 – 10/26/11, © 2011 Internet2
TCP Testing • Recommend to do at least a 20 second test every 4 hours. (less
duration is ok for short RTT, can test more or less often) • TCP is elastic, numbers should be close to capacity unless link is
heavily used. If they drop it may be congestion related or real problem related
UDP Testing • Be careful, not elastic. If you are going to do it, pick your peers.
Set a bandwidth limit, try to mimic a UDP application (e.g. video)
Testing from others: • Encourage connectors to test to a measurement node that is
close. • Test to Internet2/MCNC etc.
85 – 10/26/11, © 2011 Internet2
Regular testing
Throughput Tests
86 – 10/26/11, © 2011 Internet2
Bad Throughput … What Next?
87 – 10/26/11, © 2011 Internet2
Expected Throughput
88 – 10/26/11, © 2011 Internet2
Use in conjunction with regular traceroute: • Workstation -> Server via your terminal
• Server -> Workstation via the web
Detect asymmetric routes
Cautionary notes: • Don’t rely on traceroute for latency times (e.g. icmp and routers/
switches) • Asymmetric routing by itself is not harmful, it’s a way of life. If
one of the routes has a performance problem then it is harmful.
89 – 10/26/11, © 2011 Internet2
Reverse Traceroute Use Case
Reverse Traceroute Results
90 – 10/26/11, © 2011 Internet2
NPAD Home
91 – 10/26/11, © 2011 Internet2
Similar to NDT, but spits out a lot more • Tries to “simulate” a network use based on current conditions.
• E.g. expectations are a latency and a speed, will tell you if this is or is not possible
Also replies on Web100, same caveats apply
“Mom Feature” … you can retrieve the HTML results from past tests. Users can mail these in trouble ticket reports.
92 – 10/26/11, © 2011 Internet2
NPAD = Network Path Diagnostics
NPAD Start – Pick some Expectations
93 – 10/26/11, © 2011 Internet2
NPAD In Progress Testing
94 – 10/26/11, © 2011 Internet2
NPAD Results
95 – 10/26/11, © 2011 Internet2
Another Interesting one
96 – 10/26/11, © 2011 Internet2
Cacti
97 – 10/26/11, © 2011 Internet2
Currently only monitoring the localhost • Memory/cpu things
• Plans to add NTP monitoring as well as some more advanced localhost things (process breakdown, etc.)
Can be used for SNMP monitoring if desired • Note the poll is defaulted to about 5 minutes
98 – 10/26/11, © 2011 Internet2
Using Cacti
Cacti Display
99 – 10/26/11, © 2011 Internet2
The term “throughput” is vague • Capacity: link speed
- Narrow Link: link with the lowest capacity along a path - Capacity of the end-‐to-‐end path = capacity of the narrow link
• U6lized bandwidth: current traffic load
• Available bandwidth: capacity – u6lized bandwidth - Tight Link: link with the least available bandwidth in a path
• Achievable bandwidth: includes protocol and host issues
Throughput? Bandwidth?
100 – 10/26/11, © 2011 Internet2
(Shaded portion shows background traffic)
45 Mbps 10 Mbps 100 Mbps 45 Mbps
Narrow Link Tight Link
source sink