Sprint ##DATA tuning (changing Home Agent)

horsessmellfunny · Posted: Sat Mar 18, 2006 1:06 pm Post subject:

wow, well that is interesting those HA'S route through those locations. I am located in Colorado, north of Colordo Springs, nad 50 miles south of Denver. Interesting. I did an update user profile, and it reset my HA'S back to the 68.28.185.76 and 68.28.177.76. However, before I did the update, the HA'S I entered originally were still there which is a good sign. Hmm, interesting how this plays out!

That profile that I have in those screens isnt even my profile. Im a little ocnfused by that one. The profile in bold is mine 'rWoltering@SprintPCS.com'. The one above it I dont even know how it got there. It does connect to that other profile as well. I have seen it listed a few times in my connection profile under that other name. I dont even know what it could be! Maybe somebody else's account is mixed in with mine? Maybe its a global account that is just there for diagnostic reasons, or possibly the call for when doing the software update. That other account also has those .12 HA's.

horsessmellfunny · Posted: Sat Mar 18, 2006 2:17 pm Post subject:

The HA's 68.28.89.76/77 provide the best possible service. After I did the update user profile, and everything defaulted, my speed was alot slower, and pages took alot longer to load. I also noticed I coudnt even PING anymore,I found that a little odd. I reverted back to the above HA'S and have been happ so far. I guess the HA's have an impact on 1xRTT as well as EVDO.

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

I'm interested in testing the HA addresses for anyone who is willing to do a traceroute and post results here or in my mailbox. This does not expose your IP address. We could develop a better understanding if HA addresses impact performance.

In Windows....
Start->Run->type... cmd
At command prompt: tracert www.evdoforums.com

To copy in command window, you have to 'mark' the area and then you can paste here.

In Mac/Linux, go to a terminal session...
traceroute www.evdoforums.com

Sample traceroute...
1 181 ms 132 ms 159 ms 68.28.83.69
2 * * * Request timed out.
3 162 ms 161 ms 184 ms 68.28.90.54
4 164 ms 131 ms 185 ms 68.28.90.6
5 * * * Request timed out.
6 * * * Request timed out.
7 165 ms 186 ms 190 ms 68.28.90.97
8 286 ms 182 ms 138 ms 68.28.90.17 (this is the line of interest, just before line with site name.)
9 173 ms 159 ms 185 ms sl-gw17-atl-1-0.sprintlink.net [160.81.107.21] (and this site name)

You can 'break' out of the traceroute once the first entry with a site name appears.

If you'd like, also perform a few performance tests at...
http://speedtest.frontiernet.net

Knowing the card model and signal strength in dBm would also help. Wave the mouse over the signal bars in Sprint Connection manager. Typically it would be -50dBm to -120dBm.

Post results along with traceroute.

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

Another curious find. When looking at the dslreports archive for Sprint, it looks like 'area2.spcsdns.net' consistently gets the highest results.

http://www.dslreports.com/archive/spcsdns.net?1=1&p=1

Very few results <500Kbps are from area2.

http://www.dslreports.com/archive/spcsdns.net?1=1&p=2

I was able to confirm via testing that the 'area' domain directly correlates to the HA address. Area 2 is primarily the HAs in Kansas City. It looks like even if a test is from either coast and the HA is set to a KC node, it performs best.

I'll be able to test this when in Florida next week and see if an 'area2' HA performs better than in one, say in Altanta. Will also try when I'm in Vegas next month.

horsessmellfunny · Posted: Wed Mar 22, 2006 1:33 pm Post subject:

Tracing route to evdoforums.com [209.242.15.83]
over a maximum of 30 hops:

1 666 ms 439 ms 459 ms 68.28.185.69
2 * * * Request timed out.
3 430 ms 584 ms 416 ms 68.28.91.54
4 534 ms 440 ms 561 ms 68.28.91.6
5 * * * Request timed out.
6 * * * Request timed out.
7 370 ms 620 ms 546 ms 68.28.91.97
8 531 ms 935 ms 378 ms 68.28.91.17
9 518 ms 400 ms 682 ms sl-gw17-kc-1-0.sprintlink.net [160.81.107.21]
10 382 ms 499 ms 400 ms sl-bb20-kc-1-2.sprintlink.net [144.232.23.57]
11 524 ms 395 ms 600 ms sl-bb25-chi-9-0.sprintlink.net [144.232.20.109]

12 438 ms 522 ms 418 ms sl-bb23-chi-15-0.sprintlink.net [144.232.26.93]

13 498 ms 420 ms 601 ms sl-gw22-chi-8-0.sprintlink.net [144.232.10.10]
14 399 ms 560 ms 400 ms sl-abovenet-21-0.sprintlink.net [144.228.207.54]

15 518 ms 419 ms 600 ms 64.124.78.21.dls.net [64.124.78.21]
16 419 ms 541 ms 435 ms g1-0-25.core.lith.dls.net [209.242.24.8]
17 514 ms 420 ms 605 ms f9-0-0.mux.lith.dls.net [209.242.14.218]
18 445 ms 520 ms 441 ms 209-242-15-81.rev.dls.net [209.242.15.81]
19 534 ms 505 ms 579 ms 209-242-15-83.rev.dls.net [209.242.15.83]

Trace complete.

This is on 1xRTT and set to the Kansas City HA's. Setting the HA to Kansas City makes a HUGE difference. Before I changed them, even 1xRTT was fairly slow. I have not had the time to head 8 miles south to an EVDO area yet because I have the flu and am confined to my bed, and the toilet. Even on 1xRTT those HA'S make a HUGE difference!

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

Your HA must be set to 68.28.89.76 or 77...

8 531 ms 935 ms 378 ms 68.28.91.17
9 518 ms 400 ms 682 ms sl-gw17-kc-1-0.sprintlink.net [160.81.107.21]

There is a pattern that the HA address is related to the line before the site name. In this case, the subnet on line 8 is 91. The pattern is that the HA is that subnet minus 2 in every case. And the node appears to always be 76 or 77.

That subnet is in KC and is in area2. It's also consistently the fastest subnet of all I've tested so far.

BTW, I have confirmed with testing that Chicago nodes are in area7 and often do not perform consistently as well as area2.

Michael, are you up for testing a Sprint card on a KC node vs. a Chicago node?

horsessmellfunny · Posted: Wed Mar 22, 2006 4:32 pm Post subject:

that is very odd because my HA's are definetly set to the Kansas City HA;s. They must be getting over ridden somewhere by Sprint.

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

In Tampa today and took about an hour to do quite a bit of testing to see how HA might make a difference.

One thing I can confirm is that the HA is tied to the account, not the location. When I connected, the HA stayed the same as home (Missouri). When I did an update profile, it kept the HA the same. If I changed the HA, and connected it would use the new HA but as soon as I performed an Update Profile, the HA would set back to the original. So the HA is tied to the account, not location.

What this says is that if you are in NYC and your account is tied to a NYC area Home Agent, when you travel to the West Coast, all of your traffic is sent to the W Coast tower, to an HA in NYC, to wherever website you are visiting, probably back to NYC HA and back to W Coast tower. It may be beneficial to set the HA to somewhere closer if traffic is consistently slower.

But then there may be some HAs that are slower than others...

Then I tried setting to an HA in Atlanta to see how it performed compared to others around the country. It was hard to do because we were driving most of the time. It was pretty clear that if you have a signal numerically higher than -100dBm, it would be difficult to get good performance - likely under 500Kbps, sometimes much worse. When getting -80dBm or better, the chances were higher to get decent performance and better signal will give a better chance of 1Mbps. Maybe not always the case because there are other factors but chances are higher.

When testing numerous times and having signal better than -80dBm, I was often able to get >1Mbps on the KC and Chicago HAs. On the Atlanta HA, I rarely was able to get >1Mbps and averaged 600Kbps. The tests weren't consistent though because sometimes we were moving. I do suspect the Atlanta HA has issues but can't say for certain yet.

Will be in Atlanta tomorrow and hope to do some more testing. I would say with a little more confidence that changing the HA may make a difference and is worth experimenting for those who are explorers.

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

Am doing a roadtrip from Tampa to St. Louis and was in Atlanta yesterday, Chattanooga today. Traffic was a bitch entirely through Georgia, averaging under 40mph.

EVDO coverage has been pretty much what the Sprint maps show but far greater coverage than map borders, 5-10 miles beyond what the maps indicate in most cases but a couple times the maps were about right.

Did some more testing in Atlanta metro with Kansas City, Chicago and Altanta HAs. Of 10 tests on Atlanta HA, only one test was over 1MB/sec and the others were 300-700Kbps. On the KC HA, 8 tests were over 1MB/sec and 2 were around 600Kbps. On the Chicago HA, most were 800K-1Mbps. Chattanooga is about the same story trying different HAs. Did the Atlanta tests on Saturday around 4-5PM EST and the Chattanooga test after 11PM. So there may be problems with the Atlanta HA even late at night.

Also noticed that dBm any worse than -95dBm will result in very poor performance. If dBm is better than -70dBm, you'll get great performance if on a higher performing HA. Even if dBm is better than -60dBm, I only get OK performance on the Atlanta HA while great performance on the KC or Chicago HAs.

I'd conclude with 90% confidence that if you have better than -70dBm but very poor performance, you are either on a poor performing HA or the tower you are talking to is heavily loaded. If the performance is still slow late at night, the chances are greater the HA is the issue rather than the tower. There are other factors too but I'd give 90% confidence that this is the case.

This is based on the S620 card. The other cards may have differen dBm behavior.

I think I'll send this thread to a Sprint engineer and see what they say about it. There are some W Coast HAs that are consistently slow too.

jsflgy · EVDO Newbie Joined: 03 Apr 2006 Posts: 6

FYI - The .76 and .77 HAs are the same box. Performance should be identical. I am surprised you find a noticable difference from one area to another, but am impressed how much research you have done so far.

In theory it should be best to leave the HA setting alone, as it should be set to the HA pair closest to your home location which would keep the FA-HA communication on a LAN vs. a WAN. In actual practice the difference is barely discernable.

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

I wondered if they were the same box. The more tests I've run, the more I suspected they might be. Do you know if the HAs are also the proxy or if the HAs direct to a proxy? The proxy that does the image compression that is. If the case, it's possible that one segment redirects differently than the other, which could explain the ocassional performance differences.

I do think that changing the HA can make a noticable difference. It was pretty clear in my tests. Also, if your HA is pointing to NYC and you are on the W Coast, you'd think there'd be some latency differences as well. I thought about doing some latency comparisons too. Am in Vegas end of next month and might try latency tests as well.

It would be interesting for others to try if getting good signal strenght but subpar performance.

I do think Sprint needs to check into some of the lower perfoming HAs, or perhaps the proxies that they may point to.

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

Good news. Looks like Sprint has addressed the HA issues. I don't know if this thread influenced the fix but I sent this thread to a Sprint engineer and another on the forums has responded. They responded and said improvements have been made and that there should be no HA issues anymore. So I checked throughout the day.

Sure enough, all HAs tested are now able to do about the max performance of the tower I'm on, which is doing well over 1Mbps, up to 1.4Mbps. I tested the poor performers... Penn, Atlanta and San Jose HAs and a few others. All can now provide >1Mbps at any time of day where beforehand I would rarely see 1Mbps and often below 600Kbps. I did at least 5 tests each and typically got 1Mbps-1.4Mbps. The limiting factor now appears to be the tower and not the HA.

There are many factors that can impact performance and it appeared that HA performance was one of them (or the proxies that they point to). Doesn't seem to be a factor anymore at this point.

Latency might still be a small factor if your HA address is assigned to an NYC HA and you traveled to LA, but that would probably add only 40-50ms at most. Relative to the 150ms-400ms that is typical with EVDO Rev 0, the latency loss is not significant but geeks might want to check it out.

Bingo · EVDO Newbie Joined: 09 Apr 2006 Posts: 5

xenophon, thanks for all the good information, I just found this thread today and I immediatly tried the ##debug command on my PC5740 card and it works too however it displays a different window.
The debug screen really helped me see what is happening when I try to connect to the EVDO network here in Chicago.
The problem that i am having is that the card is not transmitting anything, at least this is what it looks like when I look at the networking icon in the task bar, when connected to the EVDO. The debug statuses show clearly what is happening, the modem connects, authenticates, goes dormant immediatly then disconnects. Why, it's beyond me. There are no error packets and the signal is very good when I get close to the window (4 bars). I tried changing the HA addresses, same action.
I guess I'll have to call Sprint again.

xenophon · EVDO Addict Joined: 30 Aug 2005 Posts: 1932

You might have to reactivate the card. Would call Sprint. If you just got the card, you may want to consider swapping for the S620 card, which is generally considered to be better.

Bingo · EVDO Newbie Joined: 09 Apr 2006 Posts: 5

I talked to Sprint, they would check the card and then replace it if there is anything wrong.
In what way is the S620 card better than 5740? What chip does it use? I am thinking maybe the 5740 can be reflashed to support the faster speeds of 3.1 when they become available. Or maybe that's just whishfull thinking. I know this card worked with evdo here with a very low signal. I was getting 600k with a -96 -100dbm signal.