ahmed alrashid May 18, 2026 · 5:13 PM

Random Network Slowdowns After Upgrading My Switch — Days of Frustration!

Hey everyone, I recently upgraded my office's core switch from an old Cisco 2950 to a newer Cisco Catalyst 2960-X. After the upgrade, everything looked fine at first — link lights were green on all ports and I could ping every device on the network. But about an hour later, users started reporting severe, random slowdowns. File transfers that normally run at 900+ Mbps were randomly dropping to 5–20 Mbps for no apparent reason. Rebooting the affected machines temporarily fixed it, but the problem kept coming back after a few minutes. Here is what I already checked: - No physical cable damage (tested with a cable tester) - CPU and memory utilization on all servers looked normal - No error messages in Windows Event Viewer - Ping tests showed zero packet loss - The link never went down — stayed UP the entire time I ran iperf3 between two servers and got wildly inconsistent results: sometimes 940 Mbps, sometimes 18 Mbps, sometimes 300 Mbps — completely unpredictable. The strangest part is that everything looks perfectly healthy on the surface. Has anyone seen this kind of behavior before? I've been dealing with this for 3 days and it is affecting the entire office. Any help is massively appreciated.

3 Comments

khaled almansour May 18, 2026 · 5:18 PM

This is almost certainly a duplex mismatch — one of the most frustrating issues in networking because the link stays UP and pings work perfectly, but throughput is all over the place. What is happening: when you replaced the switch, some of your device NICs are set to auto-negotiate while the new switch ports may have a different default configuration. When one side negotiates full-duplex and the other falls back to half-duplex (per IEEE 802.3), the full-duplex side transmits freely while the half-duplex side detects "collisions" and backs off — killing your throughput. How to confirm and fix it: 1. On your switch, run: show interfaces GigabitEthernet1/0/X Look for the duplex line. It should say "Full-duplex, 1000Mb/s". If it says "Half-duplex", you have your culprit. 2. The safest permanent fix is to enable auto-negotiation on BOTH the switch port AND the NIC — never force one side without the other: interface GigabitEthernet1/0/1 no speed no duplex negotiation auto 3. On Windows, go to Device Manager > Network Adapter > Properties > Advanced tab and set "Speed & Duplex" to "Auto Negotiation". On Linux: ethtool -s eth0 autoneg on Apply this to all affected ports and you should see throughput jump back to normal instantly.

sara alotaibi May 18, 2026 · 5:25 PM

Great answer from Khaled. I want to add one more diagnostic step that will make this 100% certain without even running iperf3. Run "show interfaces GigabitEthernet1/0/X" on your switch and look specifically for two counters: - CRC errors and input errors: These spike hard during a duplex mismatch because the half-duplex side sends collision signals that the full-duplex side has no idea how to handle — it sees them as corrupt frames. - Late collisions: This is the smoking gun. Late collisions virtually never occur on a properly configured modern network. If you see them climbing, duplex mismatch is confirmed with near certainty. Example of output that screams duplex mismatch: 5 minute input rate 45000 bits/sec 1842 input errors, 1839 CRC 76 late collision After you fix the duplex settings, run "clear counters GigabitEthernet1/0/X" and watch for a few minutes. The error counters should flatline at zero. Also worth noting: this issue gets worse under load, which explains why everything seems fine right after a reboot (low traffic) but degrades as the office gets busy. The mismatch is always there — you just feel it more when utilization goes up. Good luck — you are very close to the fix!

ahmed alrashid May 18, 2026 · 5:58 PM

IT IS FIXED! You two are absolute legends. I ran "show interfaces" on every port connected to a server and sure enough — 4 ports were showing "Half-duplex, 1000Mb/s" while the connected machines were set to auto-negotiate. The late collision counters were climbing by thousands per minute. That was the smoking gun Sara mentioned. I applied "negotiation auto" on all affected switch ports and set the Windows NICs to "Auto Negotiation" as Khaled described. Throughput immediately jumped to a rock-solid 940 Mbps across all links. I ran iperf3 for 10 minutes straight and it never dipped below 935 Mbps. Sara's tip about the late collision counter was especially useful — it let me identify the exact 4 ports that were misconfigured in under 2 minutes instead of checking every single device on the network. Three days of pulling my hair out, solved in 20 minutes. This community is incredible. Marking as solved!