Unsolved
This post is more than 5 years old
2 Posts
0
5507
October 28th, 2008 10:00
Severe performance issues with Intel NICs on PowerEdge 1950 and Windows Server 2003
I've just completed testing Intel and Broadcom NICs in our new PowerEdge 1950 servers. The Intel NICs performed horribly, in both standalone and teamed configurations, compared to the Broadcom NICs.
Can someone help me understand why the Intel performance was so horrendous? You can read all about the testing methodology and results in detail at http://thebackroomtech.com/2008/10/27/dell-poweredge-1950-nic-teaming-test-results/ but to summarize a two port Intel gigabit team performed worse than a single Broadcom gigabit NIC, and I need to know why. My boss is not happy that I spent the extra money to add the Intel VT 1000 quad port NIC to each of our new servers, only to see how bad they performed.
Thanks
-Julie
Dev Mgr
6 Operator
•
9.3K Posts
•
3 Points
0
October 28th, 2008 11:00
thebackroomtech
2 Posts
0
October 28th, 2008 12:00
Intel Drivers downloaded from Dell:
Intel_LAN_10.5_W2K3_A00.exe
Intel_LAN_10.0_W2K3_A03.exe
Intel does not have a driver listed for Windows for this NIC. Their support site says to get the driver directly from Dell.
The Broadcom NIC was already on the image, so I can't say for sure where it originated from. In Device Manager it shows up as Broadcom as the manufacturer, with a driver version 4.2.2.0.
-Julie
pcmeiners
6 Operator
•
1.8K Posts
0
October 28th, 2008 14:00
Julie, a few comments......
Aside from a driver issue, likely the main reason,embedded network interfaces are considerably faster than add-in cards, unless incorrectly developed.
http://www.buildorbuy.org/csa.html
"showed Network Utilization at ~51-56% for the Broadcom tests, and ~16-17% for the Intel tests. Why, I’m not sure."
More important, how are the system resources being utilized to get the 51-56%,( CPU, IRQ), does the on board NI degrade the overall system. Like some wannabe raid cards, good benchmarks can be obtained by over utilizing motherboard resources at the expense of the overall system. Have seen overall system performance drop >20% due to devices over utilizing resources.
Each network interface reacts to the quality of the cable installation, and NI settings differently. Unless every link will pass CAT certification, and the interface settings are optimal, I would not count on benchmarks to much.
"...teamed configurations, compared to the Broadcom NICs."
How did you test the teaming results. Teaming is not supposed to increase throughput from one client's access, but from the multiple requests from multiple clients, though I do not doubt the Broadcoms are faster due to the direct motherboard connection. Point being, you would need to be combining the benchmark results from multiple machine accesses to have legit results.
Benchmark file sizes... Copying very large files and folders with the same file and size are not common on a network. To get more realistic results, you need to use a folder with mixed type files of an average size.