A network benchmark update

Previously, I've posted several Comcast / Xfinity's benchmarks for our CATV Internet service.  (Click on "Benchmark" topic at right.)

Lately, the IPv6 service seems pretty reliable with my system.  That was a problem for some time.  It may have been a bad interaction between the Comcast plant and my local routers, but it just went away.  It may have been a Comcast reconfiguration that helped us, but it also could have been an Asus firmware update. I am now running ASUSWRT-MERLIN firmware on the Asus RT-N66U router, which I can highly recommend as an expanded and improved version of the Asus distribution.(It cured a long-standing issue with JFFS2 overflow.)

Again, I highlight the DSL Reports speed test, which checks your "real world" network performance, including the dread "buffer bloat".  Today is Boxing Day (Dec. 26), and DSL-Reports is giving me a so-so report and a good report. Here they are, separated by half an hour:

The variability may be partly due to the speed test's different selection of test hosts.  But it may be something real about Comcast or the Internet "weather".  These tests use IPv6, it appears.

For comparison, between these two tests, Comcast's own speed test shows this:

This is the available speed within the Comcast network, which seems to be the best possible result -- not fully representative of what you experience with a random Internet connection -- even if the server is fast and well-connected. (Note that the driving distance from Branford CT to Boston is really about 144 miles, not under 50. Go figure.)

Still slogging on Internet Issues

While trying to understand why IPv6 is (still) not reliable on my Comcast / Xfinity service, I rediscovered DSL Reports.  This is a great place to go to get your technical / support questions answered for DSL or Cable connections.

I have a post in for Comcast's attention.  Meanwhile, I ran the nice speed test tool, which gave me the highest-ever speed report.

High speed, high cost: it's Comcast!

Comcast (dba Xfinity) is giving us tons of Internet speed these days, at least in our corner of Connecticut.  They gave us a new gateway that is now provides the results shown to the right.  Our current service started out at 50 Mb/s, but has more than tripled with little price increase.

That's all good, but the downside is that they are charging something like $1 per month per megabit per second. High speed is occasionally useful for downloading big files, but we could survive on a fifth of what they are giving.  I suppose a big household where 5 people are all looking at their own HD videos would need this bandwidth.  But that's not us!

The question is whether a downgrade of our "triple play" service would put us at a more reasonable price / performance point.  We'll see.

My home LAN gets used for special ham radio work, servers, and software development, so I need to set up DHCP with some assigned addresses and tailor the network in other ways.  Since Comcast has seen fit to cripple its gateway to prevent me from doing this, I've added a proper Asus RT-N66 WiFi router that runs off the Comcast unit operating in "bridge" mode.  The Asus unit is very nice, including IPv6 service, as you see above.

Internet Speed Tests

At AA6E, we are using the Comcast "Xfinity" "blast" service, which originally provided 50 Mb/s download, but now seems to have been quietly upgraded to ~100 Mb/s. Oh, we seem to have IPv6 service now, too.  Effectively using that in my home network will be a challenge. (One that I don't need at this time!)

Test with http://speedtest.comcast.net (MA server):

but Boston is not 5300 miles away!

Comcast (NJ Server):

Broadband Reports offers a more sober report http://www.dslreports.com/ (NJ):

The list price for this service is pretty steep IMO ($60/mo), but a slower service is not much cheaper.

Station Computer Upgrade

I had enough!  My creaky Intel Atom-based computer had enough oomph with Ubuntu to run fldigi, but just barely.  It couldn't handle a browser and logging program running alongside very well. So the D945GCLF motherboard supporting an Atom 230 processor and 1 GB of RAM is now surplus. (See right.) Note that it has on-board serial and parallel I/O ports - a rarity now! (I originally chose the Atom board as an experiment to see how small a processor was really necessary. I answered that question.)

Upgrading was straightforward.  I could keep the power supply, case, DVD drive, etc.  The board is an MSI B75MA-P45, which will run many current Intel chips.  (Left) I selected the "Celeron" G1610 (2.6 GHz, 2 cores) to run with a "mere" 2 GB of RAM.  Choosing the configuration was helped quite a bit by Ars Technica's "Bargain Box" System Guide. My logic was that any new system, even a "bargain" system, would be way better than what I had been using.  And that has proven correct.

The hardware change went smoothly enough.  The real work (no surprise) was building the new Ubuntu environment with needed development tools and Amateur Radio applications. It will be some weeks before we get all the way back to equilibrium.

Added: I ran the BOINC Whetstone and Dhrystone benchmarks on the G1610 and then on my "big" i7-920 machine.  For 2 cores, the G1610 gives 2768 floating point and 16928 integer MIPS per core. For 2 cores (of 4), the i7-920 delivers 2869 floating and 16577 integer per core.  So core for core, today's "budget" CPU is comparable to a premium chip of a few years ago.  Such is Moore's Law. The '920 will run 8 independent execution threads across 4 cores, giving it 2-4 times more potential in terms of throughput. In practice, that throughput is only realized when I'm cranking 8 threads of BOINC apps, which is a nice thing to do, if your tastes run to extraterrestrial intelligence, pulsars, or gravity waves.

Purists will note that Whetstone and Dhrystone aren't great benchmarks, and I'd have to agree. All I will say is, they came easily to hand, and they're better than BogoMIPS.

Raspberry/Beagle/Core Benchmarks

Measured the other day using iPython:

iPython's "timeit" function is hardly the last word in benchmarking, but it can be very useful if you're thinking about Python applications.  Raspberry Pi is very powerful compared to Arduino, say, but very slow compared to modern desktop machines.   No surprises, but I haven't seen relative benchmarks before.

Raspberry Pi - 30X more time for coffee

alliedelec.com

My Raspberry Pi is benchmarking about 30X slower than one thread (out of 8) on my desktop Core i7 processor. It's an interesting challenge to do real-time Python / wxWidgets programming on the little guy, but we're making progress.