We are managing like 350 servers, in couple of different cabinets. Sometimes for a certain period of time we lost 10% of pings and session drops with a whole cabin. It is obivous that there is no problem with servers because problem starts with all servers within the cabinet at the same time and ends again at the same time. (We are monitoring servers with a dedicated server in one those cabinets)
In addition to this we read 80 mbits of traffic which we believe is not natural. But interestingly after midnight when traffic should be around 20 mbits, we read 80 mbits when this problem occur, and in the mid day we read 80 mbits again when this problem occur when normal traffic should be around 60 mbits.
DC says that there is no problem with their router, switch etc. They say that servers have problem but when we check servers, loads are close to 0 or 0.5.
So we are stuck with this, they dont accept or even comment about this problem. We really need at least a comment about it and we believe that this is the true place for this question. Hope you even comment about it.
Update (15 of July); Now DC told me that (I dont know how stupid explanation is that) the cable MAY be connected to 100 MBit port. I dont know if any 100 Mbit ports left in any DC but that is their explanation. Here is the new graph;
Thank you very much
It is possible that one of the devices connected to your rack switch has a bad NIC that intermittently becomes chatty (and floods the segment with garbage). Alternatively, it's possible that there may be a problem with the switch in spite of what the datacentre operators say, and it might also be possible that you have a bad cable.
If possible, at the time that high packet loss occurs, check the lights on the switch. Most switches will indicate an abnormality caused by a faulty NIC sending garbled traffic (though the way in which they do this varies widely, usually the light that indicates a connection will flash or something, but in any case it will be the odd one out).
There might also be something else flooding lots of traffic; consider running tcpdump or something to see if there is a likely DoS going on. Dropping packets during network congestion is normal behaviour.