Actually I have this problem since the 5.xRC versions: I have a few x86 installations running as VPN servers serving PPTP and L2TP. The 5.x versions will start loosing packets after a while. It all seems to depend on load / traffic. The heavier the load is, the sooner the packet loss will start. It will eventually happen on all of them. If look at the console, everything looks OK and no errors are logged. The only way to fix this is to reboot.
Network cards are usually realtek or intel. It also happens in virtualized environments (Xen, virtualbox, vmware) regardless of the architecture used.
I have a server with moderate load: 3mbit/s avergate traffic. This will start loosing packets usually every 24 hours. So I have to reboot every day.
Since I can reproduce this every time, I will be happy to help you track this, just tell me what to send.
GL
ps: I have 3.x and 4.x versions running the same service, in the same environment handling average loads of 30mbit/s without this problem - so I have a basis of comparison.
Can’t say I’ve seen the same problem on v5.0 but haven’t done enough testing in v5.1 to confirm or deny what you’re seeing.
Edit: bah just realised I cut off the packet error section..
I imagine that it is related to a specific service or task, so the most common features won’t bring out this problem. I noticed this very early with the RC 5.x versions - but there were so many other problems that got fixed quickly that I didn’t pay much attention to this. I however started to see the same problem in the final releases - so I assume that casual testing won’t bring out the issue.
I just had a server that started loosing packets. I attached the interface screenshot below. Most clients already dropped off, but there are some still connected, showing 0 traffic - due to the packet loss. This is version 5.0 . RX drops seems to be high, but I have that on other installations as well - so that may not be a factor.
I also created a supout.rif file in that state of the router. I’m sending that to support.
Sorry to ask this, but do you have some actual problems with services or all the panic is just about counter counting numbers, btw in 3.x and early 4.x there was no such counter available
If it is x86 then “/system hardware set multi-cpu=no” (and reboot) is good thing to try, this will load Kernel without SMP support (similar to one that they use on RouterBoards)
Also if you write to support send them 2 supouts - “good” one and “bad” one.
There are no problems with the counters. The actual problem is that unless the load is light, routeros will start loosing packets at an increasing rate. It will eventually loose all network connectivity. This happened to me on all 5.x based x86 servers, both RC and final, but never on 4.X or 3.X
I’m only using these for DNS and VPN servers, so it might be related to those services.
Currently all servers are running fine, but I expect one to go down today, so I’ll have the supouts.
I believe this problem is not specific to x86. Although, I haven’t had the chance to try on one of my x86 boxes.
I too am having exact same issues (5.0x and above), along with packet loss, things start to go down hill.
Memory diminishes within 3,4-5 days of normal client usage, then progressively gets worse as loads increase.
A reboot always fixes it. I’m running (hotspot/RB750g)with absolute minimal services.
Frustrating. Back to 4.17
I was able to send the supouts to support today. I also found an actual error that could be related to this problem. Please see screenshot below … there could be many other factors involved, but my money is on memory leak now.
I confirm having the exact same issue…
I have couple of x86 devices, same and different configs. Same and different HW. I have the issue happening in a mixed of them while some do not have this issue…
ip route cache becomes full and router starts fainting…
