Your gigabit network, what kind of speed should you really expect, Part 1 of the Network Performance series

Over the weekend I reorganized the HomeNetworkEnabled forums. It had been six months since inception and I had a little time to analyze what was useful and what wasn't. On the useful front, I realized benchmarks of network performance are useful. With this revelation I added two forums to the LAN Speed Test forum, one for wireless speed test results, and another for wired speed test results. As I started to delve in to it a little more however, I realized some context is definitely in order. In what I can only describe as a geeky epiphany, I ran around the house collecting performance figures on all of our devices. I decided to put this together as a 3-part series. The first will explore wired speeds, the second will explore wireless speeds, and the third will look at the effect of antivirus on performance numbers.

To measure speed around the house I needed a consistent testing protocol. I loaded LST Server (available for $5) on to our repurposed TiVo. The repurposed TiVo is a robust system running an 17-2600k at 3.40GHz, a Realtek PCIe GBE nic, and Windows 7 64-bit. For the clients I used LAN Speed Test v.2.0.8 (100 packets, 1MB test packet size), which is also available as a multiple computer license for $5. It should be noted that LST Server is not necessary, the free LAN Speed Test can speed test to any SMB share. However, with LST Server, the speed tests are done directly to memory vs a share, making performance numbers more accurate. In the case of SMB clients I did have to run LAN Speed Test against their shares. Their performance numbers will be reported separately to compare apples to apples. In all cases virus scan was shut off. On higher end systems I found virus scan made little to no difference in performance numbers. On the mediocre to low-end systems I found the performance dropoff a little alarming, this will be covered in Part 3 of the series.

So with that, let's look at our chart of performance numbers around the house for Windows machines that I could use LST Server with.


I organized the above chart by writing speeds. Organized this way, you can see all Windows 7 machines have faster throughput, followed by Windows 2003 Server, and then by XP. As I was running the tests and not looking at the data as a whole, I thought it was fairly obvious to me that faster performance numbers were following the faster systems in a very direct relationship. As I look at the data now, it's definitely not the case. Performance tests between my two high-end systems were exceptional, pulling in almost 90% of the theoretical gigabit limit. The other systems followed a loose pattern with OS being the biggest factor. It would have been interesting to test different OS's on the same hardware to see what numbers came up.

On reading speeds the order changes in to a pattern with no real explanation. What was surprising to me was the Latitude D630 being bested by the much less of a system Latitude D610. In reading speeds the D610 is #2 out of all the systems.


The odd thing about all of this is CPU shouldn't be the limiting factor with network throughput. Lesser CPUs can still push fast gigabit speeds, but with the results you can definitely see it was at least a consideration if only seeing the ridiculous throughput between the two quadcores. CPU seemed to be less of a consideration for writing than OS type, but in reading I couldn't see a pattern.

I asked myself what I considered "acceptable". Even the old, slow Gateway Profile5 was still reading at 288Mbps over the network. While that's less than a third of theoretical gigabit speed, it's still fast relatively speaking when compared to wireless or 100BaseT. That machine is only using that sort of bandwidth on a very rare basis anyway.

Moving on from the Windows machines, I decided to test all of the SMB clients using LAN Speed Test v2 from our repurposed Tivo. The numbers are below.


The DS111 does all of the network backups for our network, so it's truly the bottleneck for all of our PCs, if you can really call it that. At 448Mbps, it's plenty fast for overnight backups. The Vortexbox sends media to a 100BaseT Samsung TV, so those numbers are more than acceptable. The DS109 simply monitors a few 100BaseT IP cameras and does secondary DNS work, so throughput is more than enough there as well. All SMB clients had very respectable numbers compared to the Windows boxes. An interesting test with more time would be to run the LAN Speed Test SMB share test against the Windows boxes as well to get an apples to apples test here.

In summary, the gigabit theoretical limit of 1000Mbps is just that for most systems, theoretical. The two quadcores got about as close as I've ever seen, and Windows 7 seemed to get you closer as well. Even so, the slowest gigabit speeds are still nearly 3 times as fast as 100BaseT or 450Mbps connected wireless (more to come on that in Part 2). Run some of your numbers and get them in to Wired speed results forum. In Part 2 we'll be looking at realistic wireless results.