Ok, here’s a question. I have a Win2k3 server with 2 1Gb server nics in it; 1 is an integrated Broadcom, the other is a Netgear PCI-X server nic. The Win2K3 dual processor box runs a “semi-vintage” DOS-like server app which has been “updated” to run on a Windows environment; the DOSish app has 10 DOSish fat clients. The server is AD-enabled but other than DNS has little else running on it. DHCP will be on it in the future.
The network has approximately 18 XP Pro clients of which 10 use the fat client; there is a Layer 3 24 port 10/100Mb switch with 2 1Gb uplink ports for servers, etc. The integrated Broadcom server nic is attached to 1 1Gb port at a fixed IP; the other nic is disabled. The L3 switch does not have any special services configured; it essentially is running like an L2 switch. There is a DLink DSL router which supplies DHCP addresses and MAC address filtering for those machines which can have Internet access. The IP subnet range is 192.168.0.1-255
When all clients are running and there is a lot of work being done, the fat clients frequently start hanging and have to be “disconnected” from the server (this is done through the server app itself). There are few if any Windows based issues like network drives being dropped or slow logons or such.
When I examine perfmon logs, I see that the network traffic spikes tremendously during fat client usage (often peaking at 100%). This implies to me that the 1 Gb interface is being overloaded by a “piggy” application. Since I am unable to modify the app and nor do I want to, I need to find a solution to reduce the network traffic. It appears that at about 5 fat client logged on, network performance is adequate and there aren’t too many complaints.
To test the general state of connections, I ran a test where I put up 10-12 ping connection per pc with 65K payloads per ping. I then started all of the fat clients up and started using them. The network ran ok although the fat clients would show up to 50% local interface peaks in perfmon. The server showed fairly steady state traffic (about 40+%) but since I ran the test at night by myself, I couldn’t replicate the true workload. I did leave all connections running (pings included) and came back the next AM and checked on everything. All pings were still working hard and all the fat clients were still up.
So I came up with an idea where I would split off a group of the workstations onto the second Netgear server nic, thus reducing the load on the first nic. However, I’m unsure how to actually do this. I have an L2 24 port switch 10Mb/100Mb/1000Gb which I was going to use with the Netgear nic; the Netgear nic would run on a new IP subnet range like 192.168.1.x. I would then have to somehow bridge/connect the two different subnets so the new subnet could have web access if necessary. I suspect though I would have to use a Cisco router with 2 LAN interfaces to do this in order to get to a WAN and yet route two different subnets.
Has anyone ever does this? If so, could you describe/layout what kind of gear I would need to make the bridge? Or am I going off in totally the wrong direction?