LAN Network Speed

[gyoung345]

I had a question about my LAN Network Speed. I searched google and this forum and just wanted to confirm what I think I did in terms of a test. I downloaded AIDA32 which has a network benchmarking test. I have a D-Link DI-634M (non-gigabit) router connected to a D-Link 8 port gigabit switch. The 8 ports to go various rooms in my home. 8 of the ports are actively used. Attached is a diagram of my home network.

After reading many articles I see that the theoretical maximum for a gigabit network is 125 MB/s transfer rates. Though many things contribute to bring down that number (CPU, overhead, HD read/write rates).

I ran the AIDA32 Network benchmark between 2 PCs (File Server and Main Desktop PC in diagram below) on my LAN and found that the average transfer rate was 63211 KB/s. If I convert that to MB/s that comes to 63.2 MB/s or 505.688 Mbps (converted using this tool).

1. I think this tells me that I am getting to roughly 50% of the pipe. Is this correct? Is a 63 MB/s a good transfer rate on a home network? Or should I be getting more? The Main User PC has an 80GB Intel G2 SSD, while the file server is using a 160GB WD7200rpm drive.
2. As long as two gigabit devices are talking to each other it should be taking advantage of the gigabit connection, right? For instance, if I am using the Slingbox on any machine, since the Slingbox is only capable of 100Mbps it won't use gigabit.
3. Also, from what I have read, my non-gigabit capable router shouldn't be having any impact either, as long as I am not connecting a gigabit capable device to one of the ports on the router. Is this true?
4. And finally, since I have mixed devices on my network, there really isn't any reason to upgrade to a gigabit router, is there?

 

[thiggins]

500 Mbps is typical of what you may see from Gigabit devices depending on the benchmarking method used.

The biggest factors in speed are processor speed and a Gigabit interface which connects to the CPU via PCIe. Processors really aren't a factor anymore, but the PCIe Ethernet might be. You'd have to dig into your computer specs to see what is used.

Switches make direct connections between clients at speeds that depend on the link rate of the two clients that are communicated. The slowest client in a pair determines the speed. Communication between 100 Mbps clients does not affect 1000 Mbps transfers.

No need to upgrade to a router with gigabit ports unless you need the extra gigabit ports.

 

[zzing123]

63MB/s is 504Mbps, which is actually not that bad.

Another thing Tim didn't mention is that TCP/IP itself has overhead (ACK packets, headers in all the transmission units etc). Furthermore Ethernet has a lot of collisions that should be eliminated by using a switch over a hub, but of course when a packet is broadcast then all transmission stops for all clients to process the broadcast. Even using the absolute most efficient TCP/IP stacks in efficient OS's (ie not on Windows) with a cable directly linking both Hosts, they still won't get over about 85MB/s.

Generally though 60-70MB/s is what you get on Gbit normally.

 

[bwana]

What is going on if iperf gives 500 Mbits/sec but a drag and drop of an 8 gig file gives only 2 MBs/s?

 

[thiggins]

What is going on if iperf gives 500 Mbits/sec but a drag and drop of an 8 gig file gives only 2 MBs/s?

iperf and jperf results are very sensitive to the settings used. See Measuring Network Performance - Jperf and Measuring Network Performance – Jperf and TCP, Part 2

 

[bwana]

zzing123's comment suggests that 500 Mbs/s is the norm. I can confirm this with iperf but not with a windows drag and drop. What is windows doing to slow itself down? How can I troubleshoot this?

 

[thiggins]

You're comparing applies and oranges. Network file copies have more overhead than an iperf/jperf test. If you ran an FTP test, you'd get yet another number, probably faster than drag-n-drop and slower then iperf/jperf.

 

[bwana]

tnx. i also find that different nics have different results even though they are all 'gigabit'. For example, the nforce chip in the older nf4 motherboards is slower than a marvell chip on the same motherboard. (opteron dual core at 2.7ghz) - i get ~ 6 MB/s. And both of these are slower by a factor of 3 compared to the nic on the asus p5k motherboard running a intel 6600 at 3.6 ghz.