We recently launched a new server upgrade project that involves two new Dell Poweredge 2950 servers. Our original problem was that our current servers that are around 4 years old are getting bottle-necked it seems on their Gigabit network connects between the Main Database Server and Terminal Server.
In our new server setup, we purchased an Intel Quad Port PCI-E server nic for each of the two new servers (model Intel Quad Pro VT - Dell's low-profile OEM version of the Intel Quad Pro PT). Our rep sold us on the fact that we could do 'link aggregation' on these - essentially 'teaming' the 4 ports on each server's card together to create one 'aggregate' link with one IP. This follows the 802.3ad STATIC link aggregation protocol (not Dynamic as we found out later that our new Powerconnect 2748 switch does not support).
Long story short, we've had to find out the hard way that it SEEMS like in MANY, MANY hours of my own testing and extensive research, it seems that in our case of wanting to INCREASE BANDWIDTH between the new Database Server and new Terminal Server (what I call a 'ONE to ONE Scenario'), the connection maxes out at using only 1GB (25%) of the 4GB aggregate link.
We already understood beforehand that link aggregation doesn't increase bandwidth for a SINGLE conversation. However, we were under the impression that if you had MULTIPLE conversations - even if going to the same server (from new Database to new Terminal in this case) ... that it would actually utilize the full 4gb amongst multiple connections (one connection would get 1GB .. the next 1GB .. the next 1GB .. etc).
What I've found in testing by setting up three different copy sessions in Explorer of a large 6.8GB dataset (3 file copy transfers of 6.8GB going from Database to Terminal Server all over the 4GB aggregate link) ... is that it STILL only utilizes the 25% (1GB) of the 4GB aggregate link.
What I also found out from testing someone's theory that going from the Database Server over the 4GB aggregate link to TWO different IP's on the Terminal Server, is that it will THEN use around 2GB of that link instead of just 1GB - approx 47-50% instead of 24-25% (see example below):
DATA SERVER 4GB AGG LINK (10.10.15.10) ----> TERMINAL SERVER 4GB AGG LIINK (10.10.15.20)
DATA SERVER 4GB AGG LINK (10.10.15.10) ----> TERMINAL SERVER 1GB LINK (10.10.15.30)
I should mention that I've setup our Dell Powerconnect 2748 switch and setup the 4GB AGG LINK ports as 'LAG Members' - and turned on 'tagging' for those groups as well. I have also tested though directly from server to server (in a crossover type scenario) - and still get the same results - maxing the 4GB link out at 1GB in a one to one (one IP to one IP) scenario.
My question is: Can someone confirm their own real-world scenario of 'link aggregation' only helping their server bandwidth in a 'ONE to MANY Scenario' (one to many IP's) ... in other words ... from one server to MANY workstations/servers?
We're thinking of now just going with a true 4GBe Fiber Channel card in each server and connecting them together (hopefully in a crossover type scenario assuming that works going directly from fiber card to fiber card - which our rep says he's had clients do before).
