2.5GbE Networking: Setup and 1GbE Performance Comparison
Tom Fenton details setting up and using a 2.5GbE NIC and compares its performance to a 1GbE NIC, finding "The cost is the bugaboo in the equation."
In 2012, Intel released its highly popular i210-AT 1GbE NIC, which has since become a mainstay in servers. To build upon its success, in Q4 2019 Intel announced the release of two 2.5GbE NICs: the i225-LM and i225-V.
What I found interesting about these 2.5GbE NICs is that they cost slightly less than the i210-AT 1GbE NIC. The 2.5GbE NICs are starting to find their way into systems, and I finally received a system with one in it. In this article, I will discuss setting up and using a 2.5GbE NIC, and then I'll compare its performance to that of a 1GbE NIC.
The system that I received was an Intel NUC 11 Pro. This system is designed for home, office and edge work. It contains an Intel Core i5 processor, 16GB of RAM, a 512GB M.2 SSD and, of course, a 2.5GbE NIC.
When I connected the 2.5GbE NIC, it was detected and configured automatically by Windows. Device Manager showed the 2.5GbE under the Network Adapters tab.
To get a baseline of its performance, I plugged it into my 1GbE network and used iPerf3 with a NUC 11 Computer Element (CE) as the other system. The CE only includes a 1GbE NIC. The CE had an IP address of 10.0.0.216, and the Pro had an IP address of 10.0.0.214.
I started the server on the CE by entering:
On the Pro, I entered:
Iperf3 –c 10.0.0.214 –t 10 -i 5 -f g -B10.0.0.216
This started the client system (-c) and connected to my iPerf server (10.0.0.214); iPerf ran a test for 10 seconds (-t 10), showed the results in Gb (-f g), and reported statistics every 5 seconds (-i 5). The results showed that the maximum bandwidth between the two systems was 0.95Gbps. The speed of the network was limited by the 1 Gb NIC on the CE system.
As I needed another 2.5GbE NIC against which to test the performance, I purchased a $30 CableCreations USB 3 Type-A 2.5GbE adapter.
I plugged it into the CE system, and it was detected as a Realtek 2.5GbE. The NIC driver was automatically installed for it.
It was given an IP address of 10.0.0.136.
I then connected the USB NIC to the 1GbE switch, and ran my test again by entering:
Iperf3 –c 10.0.0.214 –t 10 -i 5 -f g -B10.0.0.136
The -B tells iPerf3 to use bind to the 2.5GbE NIC whose IP address is 10.0.0.136.
The results showed that the maximum bandwidth between the two systems was 0.95Gbps. The speed of the network was limited by the 1 Gb NIC on the CE system.
Although switches that support 2.5GbE are readily available, they are more expensive than the 1GbE switches. After researching 2.5GbE switches, I purchased a MikroTic CRS305-1G-4S+IN which has one Gigabit Ethernet port and four SFP+ 10Gbps ports, as well as dual ports for power. It costs $150.
I purchased two MkroTik SFP+ 10GbE adapters, which will work with a 2.5GbE network. They were $20 each.
I now I had a 2.5GbE path between the two systems.
I reran the iPerf3 test by entering:
Iperf3 –c 10.0.0.214 –t 60 -i 5 -f g -B10.0.0.136
This time, I had 2.37Gbps between the two systems.
To see if the rate was sustainable, I ran iPerf3 on it for five minutes, then I ran iPerf over the 1GbE NIC for five minutes and repeated the cycle. Below is a graph I took of the network traffic using ControlUp for Physical Endpoints (CPE).
There is an old axiom that when selecting a product, you can only have two of the following three conveniences, but not all: fast, easy, or cheap. This notion perfectly fits my 2.5GbE network.
It was easy to set up; the 2.5GbE NICs were automatically detected and installed without any issues, and the NICs, switch, and cables just plugged in. It was fast; I achieved 2.37 Gbps transfer rate. But was it cheap?
The cost is the bugaboo in the equation. Yes, the Intel NUC came with a 2.5GbE NIC, and the USB 2.5GbE NIC only cost $30. I was also able to use an inexpensive CAT 6 wired to connect the devices together, but my switch with four SPF+ adapters cost $230, or $57.50 per 2.5GbE port. In comparison, I can buy 1GbE 8-port switches for $20, or $2.50 per 1GbE port. In other words, the 2.5GbE switch is 23 times more expensive than a 1GbE network.
The good news is that with 2.5GbE becoming more common in systems, the price of 2.5GbE switches is starting to fall rapidly. In fact, QNAP recently released QSW-1105-5T5, a 5-port 2.5GbE switch that does not require SPF+ adapters. It is listing for $137, or about $27 per port -- only 10 times as much as a 1GbE switch. So, the question remains: to get two and a half times the performance, is it worth 10 times as much in cost?
Tom Fenton has a wealth of hands-on IT experience gained over the past 25 years in a variety of technologies, with the past 15 years focusing on virtualization and storage. He previously worked at VMware as a Senior Course Developer, Solutions Engineer, and in the Competitive Marketing group. He has also worked as a Senior Validation Engineer with The Taneja Group, where he headed the Validation Service Lab and was instrumental in starting up its vSphere Virtual Volumes practice. He's on Twitter @vDoppler.