Test and Tool Reference¶
Introduction¶
Everything in pScheduler begins with requesting a test which in turn will be carried-out by a tool. When you run the pscheduler task command the first required argument is the test type and all arguments that follow are dependent on the type of test. The document aims to give you a list of tests and their corresponding tools provided by a default pScheduler installation on perfSONAR. It also tries to give an idea of the options available in each and other important details.
Note
This document highlights the list of test types and tools installed by default when installing pScheduler with perfSONAR. It is possible for others to write their own test and tool plug-ins so this list may not be exhaustive of all tests and tools available.
Test Types¶
clock¶
- Description:
Measure the clock difference between hosts
- Tools:
psclock
disk-to-disk¶
- Description:
Network testing of throughput and Read/Write speeds
- Tools:
curl, globus
dns¶
- Description:
Measure DNS transaction time
- Tools:
dnspy
http¶
- Description:
Measure HTTP response time
- Tools:
curl
latency¶
- Description:
Measure one-way latency and associated statistics between hosts
- Tools:
owping
latencybg¶
- Description:
Continuously measure one-way latency and associated statistics between hosts and report back results periodically
- Tools:
powstream
rtt¶
- Description:
Measure the round trip time and related statistics between hosts
- Tools:
ping
simplestream¶
- Description:
Used primarily for testing, sends a simple “HELLO” message between two hosts using TCP
- Tools:
simplestreamer
throughput¶
- Description:
A test to measure the observed speed of a data transfer and associated statistics between two endpoints
- Tools:
iperf3, iperf2, nuttcp
s3throughput¶
- Description:
A test to measure the throughput of S3 web service storage
- Tools:
s3-benchmark
trace¶
- Description:
Trace the path between IP hosts
- Tools:
traceroute, tracepath, paris-traceroute
Test Classifications¶
Each test classified in one of four categories that determines what other tests can be run in parallel:
Exclusive - An example is a throughput task. If you have very little whitespace in this category then you may have difficulty finding a timeslot for new tests. These test can run in parallel only with background tests.
Normal - An example is a task with a test type of latency.
Background Single-Result - Background tests that produce single result. Example test types include rtt and trace or clock. These tests can run in parallel with anything else.
Background Multi-Result - Background tests that produce multiple (streaming) results. Example test type is latencybg. It is not uncommon to have this column look almost entirely solid if you have latencybg tasks since they run continuously. These tests can run in parallel with anything else.
Non-Starting - These are runs that could not find a time-slot. A very important note, and common point of confusion, is that the time shown is the earliest possible time in the slot it was trying to schedule. This IS NOT the time when the scheduler tried to find a slot, failed and labelled it as a non-start. pScheduler uses a schedule horizon so likely attempted to schedule the run 24 hours in advance. A large number of runs in this category may be the indication of a busy host where it is difficult for exclusive tasks to find a timeslot.
Preempted - These runs were preempted by another with higher priority.
You may visualize different types of tests using pscheduler plot-schedule command. See also Visualizing the Schedule with pscheduler plot-schedule.
You can know which scheduling class a test belongs to by running the pscheduler plugins tests command.
throughput Tests¶
Throughput tests support the following arguments:
pscheduler task throughput --help
Usage: task [task-options] throughput [test-options]
-h, --help show this help message and exit
-s SOURCE, --source=SOURCE
Sending host
--source-node=SOURCE_NODE
pScheduler node on sending host, if different
-d DESTINATION, --dest=DESTINATION, --destination=DESTINATION
Receiving host
--dest-node=DEST_NODE
pScheduler node on receiving host, if different
-t DURATION, --duration=DURATION
Total runtime of test
-i INTERVAL, --interval=INTERVAL
How often to report results (internally, results still
reported in aggregate at end)
--link-rtt=LINK_RTT Approximate link round-trip time (ISO8601 or integer
ms)
-P PARALLEL, --parallel=PARALLEL
How many parallel streams to run during the test
-u, --udp Use UDP instead of TCP testing
-b BANDWIDTH, --bandwidth=BANDWIDTH
Bandwidth to rate limit the test to, supports SI
notation such as 1G
--bandwidth-strict Never go faster than --bandwidth, even to make up for
lost time.
--burst-size=BURST_SIZE
Limit bursts of packets to this number
-w WINDOW_SIZE, --window-size=WINDOW_SIZE
TCP window size to use for the test, supports
SI notation such as 64M
-m MSS, --mss=MSS TCP maximum segment size
-l BUFFER_LENGTH, --buffer-length=BUFFER_LENGTH
length of the buffer to read/write from
--ip-tos=IP_TOS IP type-of-service octet (integer)
--ip-version=IP_VERSION
Specify which IP version to use, 4 or 6
-B LOCAL_ADDRESS, --local-address=LOCAL_ADDRESS
Use this as a local address for control and tests
-O OMIT, --omit=OMIT Number of seconds to omit from the start of the test
--no-delay Set TCP no-delay flag, disables Nagle's algorithm
--congestion=CONGESTION
Set TCP congestion control algorithm
--zero-copy Use a 'zero copy' method of sending data
--flow-label=FLOW_LABEL
Set the IPv6 flow label, implies --ip-version 6
--client-cpu-affinity=CLIENT_CPU_AFFINITY
Set's the sending side's CPU affinity
--server-cpu-affinity=SERVER_CPU_AFFINITY
Set's the receiving's side's CPU affinity
--single-ended Run a test directly to a host without pscheduler.
--single-ended-port=SINGLE_ENDED_PORT
Run a test directly to a given port.
--reverse Reverses the direction of the test.
--reverse-connections
Make connections from destination to source where
possible.
--loopback Run both client and server side in a loopback test.
The currently supported throughput tools are iperf2, iperf3, and nuttcp. iperf3 is the default. Note that not every tool supports every option. The following table summarizes tool specific option. Other pScheduler options are supported by all tools.
option |
iperf2 |
iperf3 |
nuttcp |
|---|---|---|---|
–omit |
X |
||
–congestion |
X |
X |
|
–zero-copy |
X |
latency Tests¶
Latency tests support the following arguments:
pscheduler task latency --help
Usage: task [task-options] latency [test-options]
-h, --help show this help message and exit
-s SOURCE, --source=SOURCE
The address of the entity sending packets in this test
--source-node=SOURCE_NODE
The address of the source pScheduler node, if different
-d DEST, --dest=DEST The address of the entity receiving packets in this test
--dest-node=DEST_NODE
The address of the destination pScheduler node, if different
-c PACKET_COUNT, --packet-count=PACKET_COUNT
The number of packets to send
-i PACKET_INTERVAL, --packet-interval=PACKET_INTERVAL
The number of seconds to delay between sending packets
-L PACKET_TIMEOUT, --packet-timeout=PACKET_TIMEOUT
The number of seconds to wait before declaring a
packet lost
-p PACKET_PADDING, --packet-padding=PACKET_PADDING
The size of padding to add to the packet in bytes
-C CTRL_PORT, --ctrl-port=CTRL_PORT
The port to use for making a control connection to the
side acting as a server.
-P DATA_PORTS, --data-ports=DATA_PORTS
The port range to use on the side of the test running
the client. At least two ports required.
-T IP_TOS, --ip-tos=IP_TOS
The port range to use on the side of the test running
the client. At least two ports required.
--ip-version=IP_VERSION
Force an IP version when performing the test. Useful
when specifying hostnames as source or dest that may
map to both IPv4 and IPv6 addresses.
-b BUCKET_WIDTH, --bucket-width=BUCKET_WIDTH
The bin size to use for histogram calculations. This
value is divided into the result as reported in
seconds and truncated to the nearest 2 decimal places.
-f, --flip In multi-participant mode, have the dest start the
client and request a reverse test. Useful in some
firewall and NAT environments.
-r, --reverse Report results in the reverse direction (destination
to source) if possible.
--traverse-nat Take steps to traverse outbound NAT
-R, --output-raw Output individual packet statistics. This will
substantially increase the size of a successful
result.
The currently supported latency tools are owping (used by default) and twping. When using twping the destination (-d) can be any network device acting as a TWAMP Server and Session-Reflector, see RFC-5357 for more details (TWAMP Light is currently not supported by perfSONAR).
rtt Tests¶
RTT tests support the following arguments:
pscheduler task rtt --help
Usage: task [task-options] rtt [test-options]
-h, --help show this help message and exit
-c COUNT, --count=COUNT
Test count
-d DEST, --dest=DEST Destination host
--flow-label=FLOW_LABEL
Flow label
--fragment Allow packet fragmentation
--no-fragment Don't allow packet fragmentation
--hostnames Look up hostnames from IPs
--no-hostnames Don't look up hostnames from IPs
-i INTERVAL, --interval=INTERVAL
Time to wait between packets sent
--ip-version=IP_VERSION
IP version to use
-s SOURCE, --source=SOURCE
Source address or interface
--source-node=SOURCE_NODE
Source pScheduler node, if different
--suppress-loopback Suppress multicast loopback
--no-suppress-loopback
Don't suppress multicast loopback
--ip-tos=IP_TOS IP type-of-service octet (integer)
--length=LENGTH Packet length
--ttl=TTL Time to live
--deadline=DEADLINE Deadline for all measurements to complete
--timeout=TIMEOUT Timeout for each round trip
--protocol=PROTOCOL Protocol used to measure round trip time
The currently 2 supported protocols for RTT measurements are icmp and twamp. When using twamp you need to make sure that the destination (-d) is a TWAMP Server and Session-Reflector, see RFC-5357 for more details (TWAMP Light is currently not supported by perfSONAR).
trace Tests¶
Trace tests support the following arguments:
pscheduler task trace --help
Usage: task [task-options] trace [test-options]
-h, --help show this help message and exit
--algorithm=ALGORITHM Trace algorithm
--as Find AS for each hop
--no-as Don't find AS for each hop
-d DEST, --dest=DEST Destination host
--ip-version=IPVERSION IP Version
--length=LENGTH Packet length
--probe-type=PROBETYPE Probe type
--fragment Allow fragmentation
--no-fragment Don't allow fragmentation
--first-ttl=FIRSTTTL First TTL value
-s SOURCE, --source=SOURCE Source address
--source-node=SOURCE_NODE Source address
--hops=HOPS Maximum number of hops
--queries=QUERIES Queries sent per hop
--hostnames Resolve IPs to host names
--no-hostnames Don't resolve IPs to host names
--dest-port=DESTPORT Destination port
--wait=WAIT Wait time
--sendwait=SENDWAIT Wait time between probes
--ip-tos=IP_TOS IP type-of-service octet (integer)
The currently supported trace tools are traceroute, tracepath, paris-traceroute. traceroute is the default.
Note
Please note that if you have a server that has more then one network interface the tracepath tool does not provide an option to select the outgoing source interface.
http Tests¶
HTTP tests support the following arguments:
pscheduler task http --help
Usage: task [task-options] http [test-options]
-h, --help show this help message and exit
--url=URL URL to query
--header=HEADERS HTTP header for request, format 'HeaderName: Value'.
May be repeated.
--parse=PARSE String to parse for
--host=HOST Host to run the test
--host-node=HOST_NODE Host to run the test
--ip-version=IP_VERSION
Specificy which IP version to use, 4 or 6
--always-succeed Treat HTTP failures as successes, regardless of
response code
--keep-content=KEEP_CONTENT
Amount of content to keep in bytes; default none, 0
for no limit, SI units supported
--timeout=TIMEOUT Timeout for each query attempt
dns Tests¶
DNS tests support the following arguments:
pscheduler task dns --help
Usage: task [task-options] dns [test-options]
-h, --help show this help message and exit
--host=HOST Host to run the test
--host-node=HOST_NODE Host to run the test
--nameserver=NAMESERVER Nameserver to query
--record=RECORD Record type to query (One of a, aaaa, ns, cname, soa, ptr, mx and txt)
--query=QUERY String to query
--timeout=TIMEOUT Timeout for each query attempt