Hi
My customer has a five node Windows 2008R2 NLB cluster (running Exchange 2010 CAS servers). The cluster is single NIC, running multicast and is virtualized in VMware. There are around 8000-9000 users connected, mainly Outlook but also ActiveSync and OWA connections
Often on a Monday morning these servers seem to suffer from network problems. Usually one server at a time is affected, and is either not pingable or with large packet loss. Clients are disconnected and the server event logs report issues connecting to domain controllers, DNS servers etc. The server appears to recover but the symptoms either reoccur or move around to another server, until the whole NLB farm is rebooted.
We had this issue last year, always on the first working day of the week until a 5th node was added to the NLB cluster which resolved the issue until now.
One thing we have discovered is when the network problems start the performance monitor counter "packets received discarded" rises sharply on the affected servers. In the most recent occurrence, this counter rose to 3.3 million discarded packets over a couple of hours. At the same time the packets received and transmitted counters fall nearly to zero.
This looks like a load issue to us. CPU and memory looks normal so is there a limit to the number of connections NLB can handle? Are there any other performance counts we should be looking at to resolve this issue?
regards
Nick