Battleangel | Personal

In a previous post, while generally griping over hardware, I happened to mention that I’m looking into NAS enclosures and that I’m stuck on 100BASE-T Fast Ethernet and would like to move to 1000BASE-T Gigabit Ethernet. The snag being that I already have a router (D-Link DIR-635) that works perfectly well (first ever D-Link router that did) and that upgrading to lets say D-Link DIR-655 will be really expensive and feel extremely redundant. So my plan was to wait for the right opportunity and run 100BASE in the meantime.

Another concern was twisted pair cabling. While a prospective NAS would most likely sit comfortably next to this computer since this is where the router must be, and still another computer is stuck on 802.11n, I also have a media computer in a small form factor case in another room. There is about 20 – 25 m alcatel branded cat 5e cabling passing through walls and ceilings in-between. Replacing it doesn’t involve much of a cost (if DIY) but may incur bodily harm climbing a much too short and unstable ladder to get to the attic, from the outside. So I was hoping I could continue to use this cabling setup.

To this end I proceeded with some preliminary benchmarks. I connected the cat 5e cabling directly between the two computers in question and started copying files over the windows network using Teracopy. Not much of a benchmark tool, but it does show the transfer speed all through the process. Results were extremely underwhelming. Transferring from MEDIAPC to MAIN (this computer) crawled along at 13-36 MB/s. Average was probably in the low 20s at best. Transferring from MAIN to MEDIAPC yielded a more encouraging 27-43 MB/s with an average of 30+. Both computers have the same type of NIC (onboard Realtek PCIe GBE Family Controller), both have modern fast CPUs and both have relatively fast, but truecrypt protected hard drives. For a while I thought the CPU utilization of truecrypt could be pushing the dual core MEDIAPC over the edge. But although it is true that terminal does show double digit CPU usage for disk intensive operations (unpacking win7 iso is 30% or so there, 5% here), LAN transfers usually stress the receiving computer the most. I don’t know what to make of this but it’s all besides the point. The point is that I got nowhere close to what real world gigabit speeds should amount to. The bloody NAS can do 60 MB/s or so in the real world and that was what I was hoping for.

I proceeded with some NTttcp tests. 100Mbit/s tests interestingly showed 12.5 – 17.3 % CPU utilization for MEDIAPC and 5 – 6.5 % for MAIN while sending / receiving data respectively. NTttcp also shows 94 / 100 Mbit utilization which is something I certainly don’t get while moving files in the real world over a microsoft network. But I suspect that these other benchmark tools that I’ve tried, NTttcp, Aida32 and Passmark don’t bother with reading to much to and from disk. And why would they, if you task the software with doing a network benchmark, wouldn’t you expect it to test throughput and not other bottlenecks. Here is what I’ve found, including results from both NTttcp and LAN Speed Test.

The LST scores are really depressing, but they correspond well with what I experienced using Teracopy. In other words, real world results. Obviously I wasn’t expecting to get 900 Mbit/s outside of theoretical tests. But 104 Mbit/s is pretty dismal. CPU data isn’t represented in the chart, but indicated about 26% on this Q6600 quad core system. And 50% on the E8400 duo on the mediapc side. Reading or writing didn’t make much difference here. Other tests, which also aren’t represented here, showed a max throughput of more like 500-600 Mbit and even at that speed pushed CPU utilization to 70% average on the duo. Add to this the stress of writing and reading back from an encrypted drive and can see that we have a problem.

Clearly something has become a bottleneck here. Overhead? Hard drives? CPU? Truecypt? Drivers? Lack of router/switch? It’s worth spending another couple of days investigating. But the point is that all the theoretical test show the same thing. This cat 5e cable can easily handle gigabit speeds. And that is what I wanted to know.