This series of playbooks accomplishes a number of performance test tasks in the network. They are briefly described below, with each one becoming more time consuming and comprehensive than the one before it.
-
sperf: Short-term performance test that builds a matrix of average RTT times. This is suitable for general public distribution. Unlike other tests, the number of columns is dynamic based on the target list. -
mperf: Mid-term performance test that builds a matrix of data from many subcomponents, listed below. It is provided to engineers only.- IP SLA udp-jitter to capture detailed performance metrics
- ICMP ping to probe for MTU/fragmentation problems
- MPLS ping (LSPV) to test MPLS label switch path health
-
lperf: Long-term performance test that focuses on detailed performance data collected over hours. It is provided in support of sustained efforts or as proof to your SP that they are not meeting their SLAs.
Given these reachability results, some basic conclusions about the state of the network are drawn and summarized in a synopsis for each target. Collected information is rolled up in a master CSV file for quick reference. Again, the length and composition of each CSV file differs between the 3 tests.
Contact information:
Email: njrusmc@gmail.com
Twitter: @nickrusso42518
Cisco IOS routers are supported today. The routers need to be on a version
that supports the exec command set ip sla udp-jitter and ping mpls ipv4
in order to function with the mperf and lperf playbooks. All three
playbooks require basic ICMP ping functionality, which all Cisco IOS
routers support.
Testing was conducted on the following platforms and versions:
- Cisco CSR1000v, version 16.07.01a, running in AWS
All Cisco IOS MPLS routers in the perf_routers group are in scope. Normally,
the playbook runs what is effectively a full IP/MPLS reachability test in the
global table. For a more rapid test, you can modify the targets list to
include only some routers, while also changing the hosts in scope. A
combination of approaches are also supported. For example, maybe logging into
one router per region then targeting 2 routers in every other region
provides a sufficient picture of the current network performance.
The following are REQUIRED for this playbooks to work:
- All routers must be in a group containing the string
"region" - All routers must be in exactly one of these regional groups
- All regional groups must be children of
perf_routers - Each regional group must have a corresponding
regional_sladict. This provides the SLA committments to all other regions, including its own (intra-regional SLA).
# hosts.yml
---
all:
hosts:
localhost:
children:
perf_routers:
children:
usa_region: # AWS North Virginia, USA
hosts:
csr1:
csr2:
emear_region: # AWS Frankfurt, Germany
hosts:
csr3:
...
Below are two examples of group variables files, one per region.
# group_vars/usa_region.yml
---
regional_sla:
usa_region: 30
emear_region: 88 # Aggressive SLA chosen to show over threshold counters
...
# group_vars/emear_region.yml
---
regional_sla:
usa_region: 120 # Conservative SLA chosen to show good performance
emear_region: 40
...`
These playbooks rely only on group_vars that are defined for the general
router group. The main variable is a sequence called targets that lists the
global hostname of any remote host. Nodes not yet activated/configured
should still be added to the list; the playbook has the intelligence to
only consider "online" targets based on /32 FIB entries present.
To save processing time, this collection of "online" targets is only run
on one host then referenced from the rest. The output parsing
is accomplished using the intersect_block custom filter.
Regional SLA values were discussed above as they are more relevant to hosts (and their geographic locations) than they are to general playbook operation.
As a minor mperf option, operators can modify the IP SLA repeat count.
For quick reachability testing, the repeat count should be small. For
reliable performance testing, the repeat count should be high. Note that
probes that run too long or incur too much loss will simply fail to provide
any results, which is a shortcoming of the IP SLA feature when invoked
from the exec shell.
The state variable is only relevant for long-term probes since the
lperf_put playbook will add or remove them, depending. The time_hrs
variable is the number of hours the probe should run and must be between 1-24.
The state can be present, absent, or restarted only.
Only the most recent complete hour of statistics is available
when long-term performance metrics are retrieved. This is a limitation
of the current playbook implementation, and may be improved in a
future revision. The repeat count is only valid for the
mperf playbook.
One final note: The targets in this list must be loopback0 IP addresses. This limitation may seem arbitrary, but it simplifies the code and generally makes sense, since we are testing reachability of MPLS LSPs in many cases. These IPs may not be the same IPs as those used for management of a given node as specified in the inventory. For networks where devices are managed in-band by their loopback0 IP addresses, the target name and the inventory name can be the same.
---
state: present
time_hrs: 1
repeat: 10
targets:
- id: 100101
target: csr1_lb0
- id: 100102
target: csr2_lb0
...
Finally, there is a common variable called scp to assist with copying
rollups off the control machine onto an SCP server:
user: The SCP username that can write files to the SCP server.host: The SCP server FQDN/IP address. Under its root directory, a directory calledperf/should be created. This is where the archives generated by this playbook are copied.
The templates are NOT commented for readability in this playbook since
they are not used for issuing commands to a network device. Rather, the
template is used for transforming IOS output into a human-readable CSV file
for further analysis and archival. Many custom filters are used and are
individually documented in the Python source code, which can be
found in the plugins/filter/filter.py file.
The templates should not be changed at the operator level.
The table below provides the LSPV codes that may appear in the mperf
sheets when testing MPLS reachability.
+------+-----------------------------------------------------------------------+
| CODE | MEANING |
+------+-----------------------------------------------------------------------+
| ! | success |
| Q | request not sent |
| . | timeout |
| L | labeled output interface |
| B | unlabeled output interface |
| D | DS Map mismatch |
| F | no FEC mapping |
| f | FEC mismatch |
| M | malformed request |
| m | unsupported tlvs |
| N | no label entry |
| P | no rx intf label prot |
| p | premature termination of LSP |
| R | transit router |
| I | unknown upstream index |
| l | Label switched with FEC change |
| d | see DDMAP for return code |
| X | unknown return code |
| x | return code 0 |
+------+-----------------------------------------------------------------------+
The table below provides the detailed explanations for the abbreviated
column headers in the spreadsheet. This is used only for mperf and lperf
playbooks as sperf only shows average RTT for brevity.
+---------------+-------------------------------------------------------------+
| COLUMN NAME | DETAILED EXPLANATION |
+---------------+-------------------------------------------------------------+
| rtt_cnt | Number Of RTT received |
| rtt_min | RTT Minimum (ms) |
| rtt_avg | RTT Average (ms) |
| rtt_max | RTT Maximum (ms) |
| rtt_ovthr | Number Of RTT Over Threshold |
| rtt_ovthp | Percentage Of RTT Over Threshold |
| lat_cnt | Number of Latency one-way Samples |
| lat_sd_min | Source to Destination Latency one way Minimum (ms) |
| lat_sd_avg | Source to Destination Latency one way Average (ms) |
| lat_sd_max | Source to Destination Latency one way Maximum (ms) |
| lat_ds_min | Destination to Source Latency one way Minimum (ms) |
| lat_ds_avg | Destination to Source Latency one way Average (ms) |
| lat_ds_max | Destination to Source Latency one way Maximum (ms) |
| jit_sd_cnt | Number of Source to Destination Jitter Samples |
| jit_sd_min | Source to Destination Jitter Minimum (ms) |
| jit_sd_avg | Source to Destination Jitter Average (ms) |
| jit_sd_max | Source to Destination Jitter Maximum (ms) |
| jit_ds_cnt | Number of Destination to Source Jitter Samples |
| jit_ds_min | Destination to Source Jitter Minimum (ms) |
| jit_ds_avg | Destination to Source Jitter Average (ms) |
| jit_ds_max | Destination to Source Jitter Maximum (ms) |
| los_sd | Loss Source to Destination |
| los_sd_per | Source to Destination Loss Periods Number |
| los_sd_pmin | Source to Destination Loss Period Length Minimum (ms) |
| los_sd_pmax | Source to Destination Loss Period Length Maximum (ms) |
| los_sd_imin | Source to Destination Inter Loss Period Length Minimum (ms) |
| los_sd_imax | Source to Destination Inter Loss Period Length Maximum (ms) |
| los_ds | Loss Destination to Source |
| los_ds_per | Destination to Source Loss Periods Number |
| los_ds_pmin | Destination to Source Loss Period Length Minimum (ms) |
| los_ds_pmax | Destination to Source Loss Period Length Maximum (ms) |
| los_ds_imin | Destination to Source Inter Loss Period Length Minimum (ms) |
| los_ds_imax | Destination to Source Inter Loss Period Length Maximum (ms) |
| pkt_ooseq | Packets Out Of Sequence |
| pkt_tdrop | Packets Tail Dropped |
| pkt_late | Packets Arrived Late |
| pkt_skip | Packets Skipped |
| voc_mos_min | Mean Opinion Score (MOS) Minimum: 0.00-5.00 |
| voc_mos_max | Mean Opinion Score (MOS) Maximum: 0.00-5.00 |
| voc_icpif_min | Calculated Planning Impairment Factor (ICPIF) Minimum: 1-20 |
| voc_icpif_max | Calculated Planning Impairment Factor (ICPIF) Maximum: 1-20 |
+---------------+-------------------------------------------------------------+