Saturday, June 27, 2009

Nifty New Intranet Speed Test

Wow
LUS has launched a nifty new intranet speed test page. It tests the speed of the intranet portion of LUS' internet offering. (And you can only get to it if you are already on the network.) The decision to treat all of Lafayette as a "campus" to make the full speed of the local network available to all subscribers—regardless of what they pay—is probably the most unique and impressive aspect of LUS' service. It results in a single very high speed community within Lafayette of 100 mbps of service. Whether you buy into the lowest speed package or the highest one; whether you are the mayor or plain Joe Citizen you get 100 mbps to talk to your fellows on the network. That's something to be proud of both technically and socially...Campus networks are typically something you can only find within large college campuses or the "campus" of large corporations like Microsoft.

That 100 mbps is the technical limit of the hardware currently in use (as I understand it) and techy types here have always been curious as to how close LUS can get to that limit. For instance for 100 mbps "fast" etherenet—ethernet being the usual reference standard for networking—is theoretically capable of 100 mbps but in real-world situations achieving 80 mbps consistently is considered good by the technical sorts that administer these things.

On that score LUS must be working with some good engineers...I got 94 mbps out of my connection on this test:


What's more its rock-steady...look at the tiny variations in the blue speed line over the test:


But the most surprising part of the above speed graph is that inconspicuous red line right at the bottom...1 ms of "delay" aka "latency." That's every bit and maybe more surprising than getting so close to the 100 mbps barrier. Latency is crucial in making next-generation interactive audio and visual applications work well. If you want to actually talk to and see someone in real time it is crucial—and is seperate from simple "speed" which might better be described for these purposes as "capacity." You need the transit time from you to the person you are talking to and back to you to be as low as possible. You do need enough speed/capacity for good video resolution and audio; but you also need a very quick response--you need low latency to make the whole experience worthwhile. (You've recall those nice clear pictures of on-scene reporters from the other side of the world talking to show's anchor. You also recall those long pauses and akward starts and stops? That's the latency part.) 1 ms of delay is astounding. Even more astounding the absolutely flat line in that graph—every point reports at 1 ms—indicates that 1 ms is simply the lower bound of this testing setup. LUS' delay varies somewhere below 1 ms. The company that designed the software clearly didn't think that it needed to ever worry about reporting delay any smaller and so is reporting all delay below 1 ms as "1 ms." LUS has confounded the expectation that delay below 1 ms isn't practical. Wow again.

So, in its summary, the software tries to tell you what your connection is good for...and in this case the decision rendered has to sound like a laconic understatment:

With 94 mbps and and at 99% consistency the service is "high enough to support a high quality" voice conversation is a vast understatement. That's enough to support, without strain due to the connection, an HD video conversation....or several. Within the network you simply won't have to worry about the network limits on what you can do. These limits are far beyond what the current hardware and software is designed to handle. —The falsely high report of 1 ms from this test software is an example of how really high speed/high quality networks expose that weakness.

Looking For A Downside
In fact that hints at the dark lining on our silver clound: We've gotten so far ahead of the curve that we are finding new choke points—choke points that few others have to worry about. In practice the most serious choke points are usually local—in the last mile network or in your ISP's regional feeder system that supplies that last mile. Server delay sometimes figures in to a slow-loading page but is usually transient. The people who run the popular servers know that slow-loading pages drives the traffic they want away and fix any issues that might arise. Even rarer is within-premise delay. Your local network has typically been so much faster than what your ISP supplies at the wall of your house that misconfigurations and out-of-date hardware don't effect your perceived speed.

But with the sorts of speeds that LUS is providing, especially on the intranet, all these formerly unimportant server issues and local network messes suddenly become the new bottleneck. For instance: I've noted before that I haven't felt obliged to upgrade my WiFi to the newer, faster N standard because I simply couldn't get enough real bandwidth from Cox for two of us to saturate my wifi's ability to push bits. That's no longer true. The 94 mbps that I got above was what I got when I connected directly to LUS' ethernet connection. When I tried the same thing through my WiFi my connection dropped to 44 mbps. I lost half of my available speed! Frankly, I'm not upset—my current WiFi hardware is set up as an a/g network. When I tested it both my wife and I had connections open. The theoretical limit of an a/g setup is 54 mbps and and the typical achieved rate is about 22 mbps. My setup is working fine. It's just old-fashioned. I need to segment the network leave my wife's old laptop connected to an a/g node which is all her 'puter can handle and connect mine to the N version. (hey! Don't look at me like that. I tried to get her a new laptop. She won't let go of the one she has.) 802.11 n is supposed to get, in practical situations, 144 mbps...plenty enough for now.

When I talked to LUS about this they said they've had a lot of issues with routers not being able to push LUS's speeds out to the laptops. This problem emerges not only in old a/g wifi routers and even some N ones but more surprisingly also over the ethernet ports in some of those routers. (Pure 10/100 ethernet routers can generally handle the speeds on wired networks, I'd presume. My wifi router, an Apple Time Machine, happily doesn't have the weakness some combined routers do but you should check yours if you use any ethernet.) So...all that speed is going to put pressure on our creaky local area networks (LANs). It's my intention to rewire my house with cat 6 wiring and install a new gig ethernet (1000 mbps) router—all our working puters can use that speed. And since I've now got the speed I'm gonna trade out the old WiFi and put in new ethernet connections to my nifty new LUS box, media computer, the newer TiVo, my PS3, and hey the TV has an ethernet port, why not? (The day is coming soon when I'll video conference on my big screen TV with folks here in Lafayette...) They'll join my printer and kid/server 'puter on the faster wired network.

So...Lafayette, the good news is that you've got a fantastic network to use—at astonishing prices too. The bad news, such as it is, is that you'll have to start paying some attention to your end of the connection for probably the first time in your life. There might be some work involved.

I'm kinda enjoying having that kind of "problem." :-) Have fun!

17 comments:

Speed said...

John,
That's awesome that they are allowing you to truly see what your speeds are instead of just guessing that it's working right. And you've brought up something that's also been an issue with people experimenting with gigabit ethernet. Tom's Hardware has an article about that explaining how most people think they will be able to FTP at gigabit speeds, but hard drive transfer rates are a new bottleneck as they can't keep up!
In my personal experience, I used to have a wireless-b router. At some point, Cox had upgraded the internet speeds to where it exceeded what wireless-b was able to do and I would get faster download speeds from my ethernet connected PC than my wireless in my laptop. After I upgraded to wireless-g, that problem went away. Because of my anticipation to upgrade to LUS fiber, the first day I saw a good deal on a wireless-n router, I upgraded so I'd be able to handle the speeds through wireless.

John said...

Hi Speed,

You were smart to move on to N in anticipation of LUS. Your point about gig ethernet is a good one. It's pretty hard to _fill_ a gig of connection between two points due to other constraints somewhere along the line. That's not to say that on a local network having a gig of _shared_ capacity won't be helpful and eliminate any issues arising from concurrent use of 100 meg ethernet...

The article in Tom's hardware was really interesting. The biggest bottleneck is the hard drive. Thanks for the pointer.
(http://www.tomshardware.com/reviews/gigabit-ethernet-bandwidth,2321.html)

Will Hinson said...

Isn't it kinda like the fox guarding the hen house? Why not have a third party test the speed like J.D. Power so it can give LUS a rating.

John said...

Will,

I think you're comparing apples to oranges in at least two places.

First, JD Power's ratings are about "customer satisfaction" not speed/capacity or latency. Customer satisfaction can vary according to a lot of things, and technical factors might well be the least of them. (PR, your line of work, ranks higher, I'm sure.) Interesting rankings for the South: http://www.jdpower.com/telecom/ratings/high-speed-internet-service-provider-ratings/south/sortcolumn-1/descending/page-#page-anchor Earthlink probably ranks above Cox because it is a chosen, not default/incumbent, service--making the comparison not entirely fair to Cox.

Secondly, The speed and latency wins I talk about in this post are intranet only...not internet. Third party comparison speed tests we see online are to a server located on the larger internet and wouldn't tell us what we want to know about the 100 mbps intranet feature of LUS' offering.

That said, I would like to see some real-world tests done between folks on the LUS network. That'd be about as "objective" as we're likely to see about intranet data and it'd be worthwhile to see. I'd caution that the testers would need to be fairly savvy sorts to get a true reading due to the sorts of factors I discuss in post and that speed brings up in his comment.

Nick Istre over at the Cajun-Asian blog and I have run a few tests between us just for fun. We came up with very similar numbers — mid 90's in speed, Latency below 1 in ms...that seemed so astonishing that I hesitated to publish. I no longer think it anomalous now that I've gotten the same figures over a less cobbled-together setup. A jog over to Geoff Daily's blog shows Nick's server logs Quality graph for the period when he changed from Cox to LUS. Pretty impressive difference—especially since Cox's figures are really pretty good.
http://www.app-rising.com/2009/04/broadband_speed_about_latency.html

Alex said...

What's funny is my old WRT54G cannot push enough bits to use my 50mbit connection (using dd-wrt i found that my router was actually using all available CPU when using ~25mbit of my 50mbit service). I'm actually making my router and using a switch.

John said...

Alex,

The WRT54G is a pretty worthy WiFi router...and the fact that it has open firmware makes it a serious geek tool. You might want to look at the various specs on Wikipedia to see which version you have.
http://en.wikipedia.org/wiki/Linksys_WRT54G_series

I'm curious to know what you mean when you say: "I'm actually making my router and using a switch."

Speed said...

John,
What kind of latency are you getting when pinging random sites on the internet like yahoo and google? With Cox internet right now I'm getting about 40ms for yahoo and 80ms for google. I'm curious to know if fiber helps those latency times.

John said...

Speed,

Well, Using speedtest.net I just got a 15ms ping to the Speakeasy server in Dallas..but a 30 ms ping to the Cox server in N.O. :-) It varies a lot by your target server and even with the target server held steady it varies over time...probably due to different paths, and different performances of nodes on the same path over time.

Not sure how fair we could get on this without fixing the route and doing multiple tests....

What do you get to Dallas on speedtest.net?

Speed said...

Yeah, it's really hard to get accurate results with ping. I was just wondering if the pings to sites on the internet seemed really low or if it's not much different.
I just did the speed test and my ping was 19ms.

LorJack said...

Speedtest.net is pretty good. A good friend of mine has been coding that for quite a while. A good ping test I use is to jump around the speakeasy network:
dfw.speakeasy.net = dallas
atl.speakeasy.net = atlanta
nyc.speakeasy.net = new york
sfo.speakeasy.net = san fran
lax.speakeasy.net = los angelas
sea.speakeasy.net = seattle
chi.speakeasy.net = chicago

Comparisons on our first hop to dallas:
Bellsouth DSL = 80ms (Lafayette is split into two ATT gateways, woe be you if you get on the "slow" gateway...welcome back to 1997 ADSL ping.) 25 ms if you are on the fast side of town. ATT routing changes at the drop of hat, so it can easily flip over to Atlanta instead of Dallas...pings fluctuate a lot.
Cox Fiber = 15ms
Cox Cable = 20ms
LUS Fiber = 10ms

LUS is the best network to be on if latency is an issue. I was hoping that LUS would have a local route to COX..but everything runs thru Dallas.

My speedtest.net varies depending on day or night testing but best was 19meg down 10meg up (I have the 10meg LUS server, so I'm happy)

Bob said...

I tried to click on your link and try your internet speed test and it's giving me a 403 forbidden error. Will try from another computer bit later, but please check it out.

John said...

Bob,

I think the problem is that you have to be on the network for it to be avaialable...are you on the LUS network. (It makes sense...it's supposed to test the intranet.)

I'm in Baton Rouge right now and I get a 403 forbidden right now over my AT&T connection.

John said...

Bob,
Back in Lafayette...the page is still up for those of us on the network....the problem must have been your location.

CLogue said...

I've had many internet connections over the years. From dial-up to Comcast and have never seen speeds of this nature. I just had LUS installed a week ago. I have several clients around Lafayette that do data back-up to my servers and the connections and speeds are rock solid. Those of us here in Lafayette are truly fortunate.

One of the best features for those of us that run home based computer businesses is that there are almost zero restrictions unlike with Cox which blocks ports. LUS blocks nothing and has a 2 terabyte per month soft cap. Amazing. Thanks LUS!

vijay said...

It's nice information....I check the my speed test here http://www.ip-details.com/ It provides ip address,ip search,domain host search,get ip details....

sanjeev said...

Theres performance monitors built right into windows (administration tools -> performance, right click on graph to add counters, then change dropdown to network and you can monitor all sorts of network info).
Mcx Tips + Free Trail

United Intranet said...

Nice post! Considerably, the article is actually the freshest on this notable topic. I agree with your conclusions and will thirstily look forward to your next updates.I will immediately grab your rss feed to stay abreast of any updates.