Network speed testing

My internet issues last week have sparked off one of those "Science Project" ideas ... I'm curious as to how fast my various networking bits go. The objective is to figure out how good the various pieces of hardware are. So - let's introduce the bits and pieces on show :

Laptop 1 - HP G70 120EA, pretty new with a 160GB SATA hard disc
Desktop - Homebrew machine, with a 250GB Seagate and 70GB WD Raptor hard discs
Router 1 - Apple Airport Express, the older model that does 802.11g only
Router 2 - Philips router working off 802.11g for wifi and Ethernet for wire
No Router - I have an Ethernet crossover cable that lets me connect 2 machines up without a router.

The laptop and desktop are working off fairly respectable parts, so any speed restrictions should be due to the networking alone.

The files I'll be transferring will be my iTunes library :-) Approximately 11GB in around 3400 files. Should be a fairly representative test, although I'd have liked there to be a better mix between Huge and ickle files. The iTunes library will be files of mostly 2MB to 8MB with not too many tiny (10 KB) or big (100 MB) files. But with the number of files to transfer being 3400, it should be fairly representative to see if there's an effect from seek time.

Total size to copy is 11.3 GB (12,202,749,873 bytes) in 3,407 files. For the calcs, bear in mind that the bytes has to be multiplied by 8 to turn it into the Bits/Second which all the sales figures use ...

If I get chance, I'll do the same test on my old laptop but that's restricted to testing with the Philips router only, as it can't do wifi networking any more. This will be where non-network speed restrictions show up as I know it has a much slower hard disc compared to the others.

Currently waiting for defragging to complete and then I can start with the heavy copying :-)

Configuration 1 : Laptop 1 to Desktop, Apple WiFi

So this is my good laptop, copying to the Desktop with its 250GB Seagate drive (has the most space) using the Apple Airport Express. Transfer link is 802.11g, the laptop is using the built in wifi, the desktop is using a Belkin 54g PCI card. Both machines are reporting "Excellent" signal quality and are claiming 54 Mbit/s transfer speeds.

Time taken to transfer the library : 3 hours 28 minutes ...

and the behaviour is as in the picture.

The 11Mbits/s is about the maximum achieved on this link but the up and down nature of the graph will tell you that it's by no means a reliable link. It also doesn't mention that streaming music to remote speakers from iTunes via the Airport doesn't work when doing a copy of this magnitude.

Sustained Data Transfer Rate = 7.46 Mbits/s - way below the 54 Mbits/s claimed speed and slower than my internet connection.

Test 2 - Desktop HD to Desktop HD

My desktop has 2 hard discs, one is a gaming install and the other is a WinXP install that's too crippled now to be viable for gaming.

Transfer time = 288 seconds
Transfer rate = 323.3 Mbits/s (40Mbyte/s)
I have good hard discs :-) Although I'm not quite sure if I believe the 323 Mbits/s, that seems too fast although they are working off independent SATA channels.

Reader hard disc = Seagate 250Gb drive, no fragmentation
Writer hard disc = Western Digital Raptor 74Gb, reasonable fragmentation

That's the best possible combination of disc to read from and disc to write to.

Test 3 - Wifi to Wifi via the Philips router
Good laptop again, trying to copy to the desktop using my old router. Case of Look Ma No Wires, with both machines connecting via wifi. I actually abandoned this test after 2 attempts because the performance was that bad. The first attempt was abandoned before I even got the chance to start the copy and the second abandoned after getting bored of seeing it occasionally hit a max of 7Mbits/s.

What this does is eliminate the Apple Airport Express as the main suspect in my poor latency in Warcraft ...

Test 4 - Laptop Wifi, Philips router, Desktop wire

This is almost certainly going to be my "gaming" configuration, when I pick up a better router :-). Laptop connects to wifi router using 802.11g and the desktop hooks up via a standard Ethernet cable. It's not a permanent fit as cable stretch leaves the router sitting in the middle of the floor ... Anyway : specs !

Laptop - wifi 802.11g claiming 54Mbits/s and Excellent signal strength
Desktop - 1Gb/s adaptor claiming a connection speed of 100Mbits/s

Copy completed in exactly 95 minutes 0 seconds (heh - give or take a second or two, one reason for the massive copy size was so that I didn't have to be ultra precise with the stopwatch :-).
5700 seconds for the transfer, giving 16.33 Mbits/s.

And as you'll see from the first network meter screenie, a much smoother transfer. There's a reasonably periodic dip in there, I'm putting that down to Vista & XP taking a ciggy break between copying one file and starting another.

Out of curiosity, I've attached a second Network Meter screenie from near the end of the copy. The way the test worked, lots of big files copied first, followed by lots of little files. The step halfway through the graph shows where the transition between big and little files came, you can see how the network performed much worse on the little files compared to the big ones.

Test 4 - Laptop to Desktop without a router

This one involved turning off the wifi adaptors on both Laptop & Desktop so they wouldn't interfere with the transfer and then hooking them together with an Ethernet Crossover cable. Normally with Ethernet nowadays, you'll have a phone type cable with separate Transmit and Receive lines. What a router/switch/hub will do is to connect one computer's Transmit to another's Receive. The more intelligent the router, switch or hub, the more efficient it is at doing that. A hub is quite dumb, it'll send all traffic to all machines. A router is smarter, it'll keep Laptop -> Desktop traffic separate from Laptop -> Speakers traffic. It will also do stuff like send Windows Live Messenger traffic to just the laptop and Warcraft traffic to just the desktop. (Network Address Translation - the outside world sees just the IP address, the router translates outside IP addresses to internal IP network addresses)

Urg - sidetracked into GeekyTalk there :-) A Crossover cable switches the Transmit & Receive around, so you can connect two machines together without using another box.

Right - Laptop has a 100Mbps adapter that's claiming 100Mbps
Desktop has a 1Gbps adapter claiming 100Mbps
(actually a different adapter as my back moans at me if I rummage in the place where the leads plug in)
Time to transfer - 24 minutes 50 seconds (1490 seconds)
Transfer rate - 62.48 Mbits/s

I've about exhausted all the testing I can do for now, save pulling out my old laptop and hooking that one up ... (would also involve reclaiming a cable from under the sofa :-) So it's time to put the results all together :

Airport Express router - wifi to wifi - 7.5 Mbits/s sustained, 11 Mbits/s max
Philips router - wifi to wifi - test failed, 8 Mbits/s max observed
Philips router - wifi to wire - 16.33 Mbits/s sustained, 18 Mbits/s max
No router - crossover cable - 62.48 Mbits/s
(worth mentioning that desktop to old laptop on this setup could only do about 33 Mbits/s)
Desktop only - hard disc to hard disc - 323 Mbits/s

I'll grab another Ethernet cable at the weekend, so I'll be able to try a Wire to Wire copy using a router :-) That'll tell me how much inefficiency comes from having a router in circuit. Even though the writing on the packet will claim 54 Mbits/s or 100 Mbits/s or even Gigabit Ethernet, you'll never get that due to protocol overhead.

Protocol overhead - splitting big files into network packets, addition of error checking and correction, wireless encryption and decryption. There's a 1 Mbit/s bus out there that will never get above 800kbits/s actual transfer, due to the operating rules it follows.

I think I've learned something and I've got a much better idea for where my poor latency in Warcraft comes from. Wifi G seems good for convenience but it's not really up to the job of chucking big files around. Wifi N would have more headroom but will still be subject to the overheads that come in through encryption.

Do I need Wifi N ? Nope. When I want the extra performance, I'll be able to use the wire hookup. Plus, as the laptop to Philips link could manage 18Mbps, that's comfortably better than my 10Mbps internet connection. So the extra speed of Wifi N would be gold plating.

Probably the last update or so for now until I buy some new bits n bobs :

This laptop to Dad's laptop via wifi - 11Mbps constant.
This laptop to Dad's laptop via wire - up to 20Mbps but a bit inconsistent.

Next steps :
New Linksys router will arrive :-) And a long cable so I can use a wire for my internet gaming ...

Addendum - I showed this post to someone at work, who got curious about the gadget in the screenies ... It's a gadget that runs in the Vista sidebar called "Network Meter", available via AddGadget.Com. Keeps my inner geek happy :-)