Network Performance Analysis and Troubleshooting
At the time of drops or slowness, we collect data from the client, the access switch, the uplink and the target system. By comparing records from the same time window, we investigate whether the problem is in the physical connection, the network or application access.
With a “the network is slow” complaint, we determine by measurement which layer the problem is in.
Diagnosis with the network’s own data
The problem can be on the client, cabling, switch port, uplink, DNS, WAN or application side. Diagnosis starts by defining the problem with reproducible criteria.
Physical
CRC, duplex, link flaps and optical values
Network
Latency, jitter, loss and route
Capacity
Uplink utilization, queue/drop counters
Implementation
DNS and service response times
What does each metric show?
Reading the right metric from the right source narrows the problem down quickly. The review is done using SNMP counters, LLDP/CDP, flow data, continuous measurement probes and, when needed, packet capture with SPAN.
| Metric | Where it comes from | What it points to |
|---|---|---|
| Interface utilization | SNMP 64-bit counters (ifHCInOctets / ifHCOutOctets) | A bottlenecked uplink, need for capacity |
| Input errors | SNMP ifInErrors | Frames received with errors. Possibility of a cable, module or port problem; physical interface counters (CRC, optical level) are reviewed together before deciding |
| Discarded output packets | SNMP ifOutDiscards | Packets discarded because they couldn’t be sent even though there was no error. Mostly points to full queues; reviewed together with the relevant queue counters |
| Latency and reachability | Round-trip time and reachability with ICMP. A suitable test method (measurement probes) for latency variation and one-way measurements | Voice and video quality, line problems |
| Packet loss | Measurement probes, interface counters | Slow file transfers, dropped calls |
| Flow data | NetFlow / sFlow / IPFIX | Shows source, destination and traffic distribution. Application recognition depends on the fields the device exports and the analysis tool |
| Neighbors and topology | LLDP / CDP | The real connection structure, undocumented devices |
We move from a symptom to the root cause
Troubleshooting is done not with random attempts but with a layered method: from the physical layer up to the application, one possibility is eliminated by measurement at each step. For intermittent problems, data is collected over several days and matched with the times of complaints.
Measurement points
To see where the problem starts, data is collected simultaneously on the client, gateway, uplink and target system.
Comparison
Normal and problem time windows are compared to isolate the bottleneck or anomaly.
Verification
The same metrics are used before and after a change; improvement is confirmed with the same measurements.
Causes you may encounter
After the problem is fixed, the root cause, the changes made and the steps to prevent recurrence are reported in writing.
Duplex mismatch
If one end is set to auto and the other to a fixed setting, it may go unnoticed at low traffic; as load increases, late collisions and error counters rise.
Saturated uplink
Nightly backups, camera recordings or large file transfers fill the uplink; users experience slowness at the same time.
Loops and broadcast storms
A wrongly connected cable can make the network unusable in a short time if loop protection isn’t configured correctly.
MTU problems
On tunnel and Metro Ethernet links, a mismatched MTU causes some applications to hang and leads to fragmentation.
What will you have at the end of the analysis?
A root cause report backed by measurement data, the fixes applied and monitoring thresholds so the problem doesn’t come back.
A documented, verifiable result
Problem timeline and scope of impact
Measurement findings and root cause analysis
Prioritized fix recommendations
Post-change verification results
Let’s clarify a detail
If your question is specific to your network, write to us; we’ll look at the current setup and give you a clear answer.
Can a one-off outage be analyzed?
Yes, if there were enough logs and telemetry at the time of the incident. If there is no data, a monitoring and logging plan is prepared to capture it if it happens again.
The problem only happens at certain times. How do you catch it?
Intermittent problems can’t be caught with a one-off check. SNMP and measurement data are collected over several days; the cause is investigated by matching complaint times with the measurements.
Do we need to add a device to the network for packet capture?
Usually not. With the switch’s port mirroring (SPAN) feature, traffic is copied to an analysis computer; the path of production traffic doesn’t change. When the destination port saturates or the switch is under load, some packet copies can go missing; so a packet not seen in a SPAN capture doesn’t mean it was lost on the network.
Is it the internet line or the internal network that’s slow; how do you tell?
Measurements are made at the same time to a point on the internal network and to fixed targets on the internet. The point where latency first appears narrows the investigation. We verify the result with target measurements, routing information and device counters. A single traceroute or ICMP result isn’t enough to conclude a carrier fault; intermediate devices may rate-limit ICMP responses.
Related network services
Problems found in the analysis are mostly fixed with the line management, device installation and monitoring work below.
Network Maintenance, Monitoring and Support Service
SNMP-based monitoring, configuration backups and scheduled maintenance.
Metro Ethernet and Internet Line Management
Links between sites, redundant internet breakouts and service provider follow-up.
Enterprise VLAN Setup and Network Segmentation
802.1Q, inter-VLAN routing, access rules and secure network segmentation.
All network services
See the full list of services.
Contact Us
Go to the request and contact form.
On the systems side, with Infrastructure Consulting and System Optimization, whether the problem originates on the server or application side is measured too.
Knowledge Center
Technical guides for this service
Network Monitoring and Maintenance Network Monitoring and Maintenance NetFlow, sFlow and IPFIX: Which Should You Choose for Traffic Analysis?
How do you choose between NetFlow, sFlow and IPFIX? We compare the difference between flow caching and packet sampling, accuracy, device load and collector sizing.9 min readRead the guide
Let’s solve your network problem with measurement
Tell us the symptoms and how long they’ve been happening; let’s agree on the scope of the first review together.