Mental model
Cisco objective 6.9 (CCNAAUTO 200-901) says “Explain the impacts of network constraints on applications”. Four properties of the network path shape what the user experiences.
| Property | What it is | Units |
|---|---|---|
| Latency | Time for one packet to go there and back | milliseconds (ms) |
| Bandwidth | How many bits per second the path can carry | Mbps / Gbps |
| Jitter | Variation in latency between packets | milliseconds (ms) |
| Packet loss | % of sent packets that never arrive | % |
An app can be fast on paper (high bandwidth) and feel broken (high latency), or vice versa. Each app has its own sensitivity profile.
Latency
Round-trip time. Dominated by physical distance (speed of light), hop count, and queueing in each router / switch.
- Local LAN: 0.1 to 1 ms.
- Cross-US (east-west coast): 60 to 80 ms.
- US to Europe: 80 to 120 ms.
- US to Asia-Pacific: 150 to 250 ms.
- Satellite GEO: 500 to 700 ms.
- Satellite LEO (Starlink): 20 to 40 ms.
Who suffers
- Interactive apps: SSH, VoIP, video calls. Noticeable above 150 ms.
- Protocols with lots of round-trips: HTTP/1.1 with no keep-alive, old SMB, NFS. Latency compounds.
- Database queries: each query is a round-trip. 10 queries at 50 ms = 500 ms of waiting.
Who does not
- Bulk file transfer with large TCP windows.
- Any app that can batch many requests into one.
Bandwidth
Capacity of the link, independent of distance.
- Home broadband: 100 Mbps to 1 Gbps.
- Office / branch: 100 Mbps to 10 Gbps.
- Data center: 10 Gbps to 400 Gbps fabric links.
Who suffers
- File transfer, backups, image downloads, video streaming at high res.
- Many concurrent users on the same link.
Who does not
- Small REST API calls (a few KB each).
- Terminal sessions.
- Chat (text).
Jitter
Variation in latency. Latency can be high but if it is CONSTANT, apps adapt. Jitter breaks that.
Who suffers
- Real-time audio / video. Jitter buffers mitigate up to a point; above 30 to 50 ms jitter, voice becomes choppy.
- Online gaming. Jitter > latency for most games.
Mitigations
- QoS marking (DSCP EF for voice) so VoIP packets jump queues on congested links.
- Jitter buffers (receive side) that smooth out small variations at the cost of a tiny extra delay.
Packet loss
Percentage of sent packets that never arrive. TCP retries them; UDP does not.
- Healthy path: 0% loss.
- Noticeable degradation: 1 to 2%.
- Severely degraded: 5%+.
Who suffers
- TCP throughput collapses on loss > 1% over long paths (TCP cuts window; recovery is slow).
- Real-time (UDP) voice / video gets artifacts because retransmits arrive too late.
- Reliable multicast (video delivery, financial feeds): one loss = many retransmits.
Common causes
- Overloaded link queues (buffer tail drop).
- Flaky physical layer (CRC errors on a damaged cable).
- Rate limiting (edge policer).
Measuring each
| Property | Tool |
|---|---|
| Latency + loss | ping <host> — look at avg + loss % |
| Latency per hop | traceroute <host> — spot the hop where latency jumps |
| Bandwidth | iperf3 -c <server> — point-to-point throughput |
| Jitter | iperf3 -u -c <server> (UDP mode reports jitter) |
| Everything, continuous | SNMP / gNMI counters per interface |
Rule of thumb for app owners
- “User says slow”:
- On a page load? Try latency + request count first; a 1-sec page with 20 requests at 50 ms RTT = 1 second overhead.
- On a file download? Bandwidth bottleneck.
- On a video call? Jitter + loss.
- “User says it works but keeps freezing”: jitter or TCP retransmits.
- “User says it fails only sometimes from one region”: inter-region latency or an upstream route issue (traceroute from there).
The CCNA Automation lens
As a network automation engineer you are often asked “why is this app slow?” and you need to:
- Identify which network property might be hurting.
- Measure it (automate the measurement).
- Report back with numbers, not guesses.
Scripts that collect ping / iperf / traceroute from many sources and compare are a staple of a NetOps CI pipeline.
FAQ
Latency vs round-trip time (RTT) vs ping — same thing? RTT = full round-trip (there and back). Latency often used loosely for the same. Ping reports RTT.
What is “bandwidth-delay product”? bandwidth × RTT. It is the number of bytes “in flight” on the wire at any moment. TCP window size needs to exceed this to achieve full bandwidth on a long path. Why default TCP windows underperform on 10 Gbps cross-country links.
What is QoS and does it fix latency? QoS prioritises certain packets through congested links. Does not add bandwidth; moves it from best-effort to priority queues. VoIP marked DSCP EF gets served first; your bulk download waits.
Which Cisco features help with these constraints? NetFlow (visibility), QoS (prioritisation), WAN optimisation (compression), IP SLA (synthetic measurements), gNMI telemetry (fast feedback). All beyond 200-901 scope but on the horizon.