Why benchmark?
Regular network snapshots, compared over long periods, show where performance deteriorates or improves. They reveal a decline before it turns into an outage and save money when faults are fixed before customers report them, reducing the load on your support team.
Anything that can go wrong will go wrong.
Four ways to set up the tests
Where the test heads are placed and what they test against determine what the results show.
Core to central instances
- Gives
- A defined set-up, precise results for this part of the network, simple planning.
- Misses
- The rest of the network.
Core mesh
- Gives
- Better coverage and insight into the quality of connections between sites.
- Misses
- Less precision; planning takes a little more effort.
Edge to central instances
- Gives
- A view close to the end customer's, with simple planning.
- Misses
- Most of the paths inside the network.
Edge mesh
- Gives
- The closest view to the end customer's, across the most communication paths.
- Misses
- Precision on individual links; planning takes more effort.
- Core site
- Edge site
- Central instance
- Test path
In practice, combine as many as you need, include end-customer devices, and extend tests end to end into the networks of connected providers.
The right test mix
Tests at full wire speed add little insight and are expensive. Small probes, sent through the network in short pulses, are enough, and they put no extra load on the network. Coast Guard therefore combines occasional short full-load tests, continuous delay variation measurements, and service tests that capture as many KPIs as possible.
| Test | How and why |
|---|---|
| Full loadRFC 2544, ITU-T Y.1564 | Lossless throughput, frame loss across rates and frame sizes, latency and bursts, with IMIX traffic to RFC 6985. Used rarely and selectively. |
| Delay variationRFC 3393 | At a low packet rate, around the clock, one way with synchronised clocks. Shows changes in the network in near real time. |
| TCPRFC 6349 | Long-running connections reveal congestion, traffic steering and problems in the core and at the edge. |
| WebDNS, TCP, HTTP | Name resolution, connection setup and throughput, against real websites or your own test servers. |
Test heads from core to customer
-
Core
At the central interconnection points. One or two 100 Gbit/s ports (QSFP28), 400 Gbit/s on request.
-
Edge
At the points of presence. One or two 10 Gbit/s ports (SFP+).
-
Pico
At the end customer. One or two 2.5 Gbit/s copper ports for network tests.
-
mosprobe-ng
For additional, parallel voice tests at the network termination point, mosprobe-ng is available with different access technologies.
At the edge and at the end customer, tests run over Ethernet or IP, with IPv4 and IPv6. Every site takes its time from GPS or a PTP grandmaster (IEEE 1588), so one-way measurements are accurate.
Control and reporting
Because high-bandwidth tests could be misused for attacks, control uses encrypted, authenticated connections over a separate management network in a secure architecture.
Central control
- Schedules hundreds of thousands of jobs a day, without overbooking the network or the test heads
- Manages firmware, profiles, configuration and status of every test head
- Combines the results of several test heads
- Scales to thousands of test heads
Web interface and database
- Job editor with weekdays, time windows, intervals, bandwidth and traffic profile
- Bulk import and export as XML for third-party planning tools
- Results can be sorted, filtered, exported as CSV or Excel, or read directly from the database
- Real-time alarms, daily and weekly reports, SLA monitoring