PERFORMANCE AND FAULT ANALYSIS

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.

We compare measurements from the same time window

With a “the network is slow” complaint, we determine by measurement which layer the problem is in.

SERVICE SCOPE / 01

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.

01

Physical

CRC, duplex, link flaps and optical values

02

Network

Latency, jitter, loss and route

03

Capacity

Uplink utilization, queue/drop counters

04

Implementation

DNS and service response times

MEASURED VALUES / 02

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.

MetricWhere it comes fromWhat it points to
Interface utilizationSNMP 64-bit counters (ifHCInOctets / ifHCOutOctets)A bottlenecked uplink, need for capacity
Input errorsSNMP ifInErrorsFrames received with errors. Possibility of a cable, module or port problem; physical interface counters (CRC, optical level) are reviewed together before deciding
Discarded output packetsSNMP ifOutDiscardsPackets 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 reachabilityRound-trip time and reachability with ICMP. A suitable test method (measurement probes) for latency variation and one-way measurementsVoice and video quality, line problems
Packet lossMeasurement probes, interface countersSlow file transfers, dropped calls
Flow dataNetFlow / sFlow / IPFIXShows source, destination and traffic distribution. Application recognition depends on the fields the device exports and the analysis tool
Neighbors and topologyLLDP / CDPThe real connection structure, undocumented devices
ENGINEERING APPROACH / 03

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.

01

Measurement points

To see where the problem starts, data is collected simultaneously on the client, gateway, uplink and target system.

02

Comparison

Normal and problem time windows are compared to isolate the bottleneck or anomaly.

03

Verification

The same metrics are used before and after a change; improvement is confirmed with the same measurements.

COMMON CAUSES / 04

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.

CONCRETE DELIVERABLES / 05

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

01

Problem timeline and scope of impact

02

Measurement findings and root cause analysis

03

Prioritized fix recommendations

04

Post-change verification results

FREQUENTLY ASKED QUESTIONS / 06

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.

01

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.

02

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.

03

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.

04

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.

ERETEK NETWORK

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.