Switch Performance Issues

Hello. I am running 2 ntp servers with 3000 requests a second each (Max connection speed). Every 6 to 8 weeks my entire internet connection crashes. I have isolated the problem to my main unmanaged switch (D-Link 105), and requires a manual reset. The switch is on a UPS system, and I run several other servers through this switch.

Does anyone have a suggestion on either an upgrade or network modification to keep the system as stable as possible, and prevent crashes in the future? Would prefer to keep running an unmanaged switch to simplify network admin.

Thank you in advance.

This could be a firmware bug if it happens on a regular basis. And as it’s an unmanaged switch you’re not even able to get some logs, so I’m afraid this might be difficult to track down. And you won’t be able to apply any firmware updates to get it fixed. Have you captured a packet stream at the moment it happens? Does your switch completely stops sending replies? Is it answering/forwarding only some packets?

Hello. Thank you for the response. No packets captured, as it only happens once every 6 to 8 weeks. All servers are unable to access the internet through the switch, not even ping gets through. The switch completely stops responding, however all lights are on, and blinking away. My http server, ntp server, routers, etc are unable to access the internet, or resolve WAN IP address, etc. When I power cycle the switch, all servers regain their internet connection.

As D-Link is a lower end switch, what are you thoughts on upgrading to a better unmanaged switch, like the Cisco SG110D-08?

Thank you.

An unmanaged switch freezing, that’s quite unusual I think. I’d try it with a different power supply before throwing it out. Maybe it’s not able to cover peaks in power consumption anymore.

When this happens, can you ping other servers in your LAN through this switch? I.e. from your server A to your server B.

Yes -> Maybe the issue is in your WAN router, and power cycling the switch also makes your WAN router restart the interface towards your switch, clearing any issue that the router may have had.
No -> Try with another unmanaged switch or with a different power supply. Most electronic devices have a limited lifespan, including switches. Maybe it’s simply broken. Getting a new five port unmanaged switch does not cost too much.

Can not ping between servers either. Think I’ll get another switch instead.

I have several WAN IP connections, but only one internet cable. I split the internet connection using my unmanaged switch so all my servers have their own WAN IPs. Don’t want to get into vLANs, etc, so the unmanaged switch is much simpler to operate.

Thanks again for the input.

Hello. I’ve been monitoring my ntp servers for the last few months, and noticed something. Before my systems become unresponsive, the number of ntp requests increases. I normally get about 3000 requests a second, and during these spikes, I see up to 15000 requests a second. I’ve got a 100 mbit up and down internet connection, so I don’t see this being a bandwidth issue. Could this be a packet flooding issue with my ISP? My internet connection starts becoming laggy, then over the next hour, becomes unresponsive. If I disconnect my two ntp servers from the network, my internet connection recovers. I’ve noticed this issue when my ntp servers are set to 1000 mbit, and 100 mbit in the ntp pool. I have reduced the connections to 10 mbit to see if this will affect the issue. The main problem in resolving this issue is that it only occurs once every 4 to 8 weeks, and I’m using an unmanaged switch.

Thank you for the input.

It could be the Fortigate problem. Those bursts can exceed 15000/sec, but only last for around 10 seconds.

Can you capture the IP address(es) causing the spikes.

Hello. Thanks for the suggestion. I placed a bandwidth limit on my servers, and seems to do the trick. They are very large limits, but ensures there is always some bandwidth for non-ntp data to get through.

What OS are your NTP servers running? If it’s linux you can use IPtables to set a rate limit (using hashlimit), to prevent each IP from querying faster than a specified rate.

Hello. Both servers are running linux. One is set up to use Wondershaper(uses IPTable), and the other is limited by a router. My set up limits the bandwidth used by my ntp servers, but does not limit the other computers connected to the network/internet.