Wierd Traces

[Dean_Henstock]

Have a look ? Sentech us IS ?

C:\>tracert www.is.co.za

Tracing route to www.is.co.za [196.35.72.14]
over a maximum of 30 hops:

1 157 ms 200 ms 96 ms 66.18.87.50
2 90 ms 156 ms 101 ms 66.18.67.106
3 138 ms 137 ms 199 ms 66.18.67.105
4 224 ms 222 ms 159 ms 66.18.67.25
5 223 ms 160 ms 156 ms 66.18.87.249
6 110 ms 140 ms 197 ms 66.18.87.241
7 133 ms 275 ms 178 ms 66.18.87.233
8 316 ms 276 ms 281 ms 66.18.87.225
9 223 ms 201 ms 157 ms 168.209.18.61
10 111 ms 279 ms 239 ms 196.26.0.8
11 190 ms 438 ms 127 ms 168.209.100.120
12 185 ms 198 ms 180 ms 168.209.217.251
13 * 163 ms 262 ms wsd1-bry.hosting.co.za [196.35.56.10]
14 353 ms 221 ms 239 ms weblive.onsite.hosting.co.za [196.35.72.14]

Trace complete.

Keep Surfing

 

[Dean_Henstock]

Ok, why does the above go through is and the below go through MTNNS

C:\>tracert www.ods.co.za

Tracing route to webserver1.ods.co.za [196.28.66.54]
over a maximum of 30 hops:

1 251 ms 182 ms 158 ms 66.18.87.50
2 109 ms 199 ms 136 ms 66.18.67.106
3 94 ms 280 ms 99 ms 66.18.67.105
4 164 ms 265 ms 238 ms 66.18.67.25
5 157 ms 179 ms 220 ms 66.18.87.249
6 205 ms 293 ms 145 ms 66.18.87.241
7 144 ms 97 ms 141 ms 66.18.87.233
8 182 ms 258 ms 439 ms 66.18.87.225
9 188 ms 220 ms 139 ms jp-ad-1--sntec-stp-1-a.mtnns.net.0.44.196.in-add
r.arpa [196.44.0.40]
10 141 ms 115 ms 242 ms jp-lpd-1--jp-ad-1-a.mtnns.net [196.44.0.33]
11 240 ms 216 ms 180 ms jinx-lpa-1--jp-lpd-1-a.mtnns.net [196.44.0.78]
12 * * * Request timed out.
13 * 155 ms 243 ms ap3660-1-S0-0-JINX.netactive.net [196.22.163.2]

14 341 ms 218 ms 261 ms ap7206-2-fe0-0.netactive.net [196.22.185.141]
15 217 ms 300 ms 259 ms 41.30.28.196.netactive.net [196.28.30.41]
16 203 ms 200 ms 261 ms webserver1.ods.co.za [196.28.66.54]

Trace complete.

Keep Surfing

 

[Dean_Henstock]

OK,

Why are the first 7 hops internal at Sentech, that has to **** up the latency ?

Keep Surfing

 

[loosecannon]

im wondering if the towers are not connected together with wireless [802.11x or simmilar] connections not microwave ... or is it possible that some nodes are on VSAT ??

im just guessing here but it could explain the latancy ...

packet loss is only explained by poor BW management it is a pity you cannot ping the gateways to determine loss on the link to the network ...

TTFN

 

[guest2013-1]

towers aren't connected in a normal spider network setup, rather similar to a daisy chain, however, if one tower goes down, some may still be unaffected, because the routing would then go through whichever tower is available. Hence why I think they're upgrading the microwave links on the tower as to cope with the demand if a tower goes down.

Dean, the reason why those two routes differ is simple, IS and MTNNS is used for peering now. It would be silly to route all local traffic through MTNNS if the servers are on IS's backbone, thus the link would go through directly to IS if the server is on their backbone, and most other traffic will travel via MTNNS.

Just keep in mind anything can change while they're fiddling with settings and stuff, and I dont pretend like I know anything for certain, these are just my opinions

Hell, my gran on a scooter with a memory stick is faster than Sentech's MyWireless!

 

[guest2013-1]

btw, looking into bandwidth management myself, I found that some (if not most) applications buffer whatever is received for you, and only pass on the data by what they decide to limit you to on their side. If the ack for the TCP packet doesn't come through in a timely fashion, or what I think happens, the buffer gets full and starts dropping packets, you would experience packet loss. Your PC then has to re-request that packet the server has dropped due to its buffer being full, and asta la pasta, you have a problem.

What I'd love to know is what significant differences (except for firmware and perhaps frequencies) there is between iburst and ipw, because why would the one offer low latency and the other not... etc

Hell, my gran on a scooter with a memory stick is faster than Sentech's MyWireless!

 

[Dean_Henstock]

So they are now peering at two points, which is a good step, the only thing is that is you tracert onthing on worldonline network it dies.

On a second note you think the would have tunneling setup on the towers so that you could not see th IP's

On a thrid note you can detect gateways with Nprobe but its linux.

Keep Surfing

 

[nonroker]

What pisses me off is the sheer amount of hops just to get out of the suckmytech network.

--
256k with 21k-intl-128k-local-hardcap(tm) technology

 

[qdada]

I am so naive.

I actually thought that each tower has a dedicated line to Sentech or IS Backbone. So; packets have to be daisy chained from one tower to the next till they reach some kind of a access point or some back bone ? Is that what you are saying ?

No wonder it sucks so much then ?

 

[regardtv]

Not too sure about noone's comments on daisy-chained towers ...

From earlier discussions and tests there does seem to be some sort of star implementation

<hr noshade size="1"><font size="1"><div align="right">8:00 Thursday 26 August 2004 is <font color="red">D-Day</font id="red">....
<s> 16 17 18 19 20 <font color="green">21 22</font id="green"> 23 24</s> 25 <font color="red"><b>26</b></font id="red">
</div id="right"></font id="size1">

 

[Dean_Henstock]

My understanding is most towers are connected to each other via microwave, some towers via diginet where line of sigh to the next tower cannot be seen, however the do bounce from tower to tower, most of the cellular providers do as well as this saves them money on the diginet they would have to hire between towers.

Keep Surfing

 

[aborg]

I have to agree with Dean and nonroker - far to many hops in the MySlothTech network - I am no network expert but surely this can't be optimal?

 

[regardtv]

While it's generally beter to reduce the number of hops a packet takes to traverse a network I think we may be misreading the trace routes a tad.

It would seem that both sides of some of the Sentech routers are responding as hops ... ie here and there 2 hops are actually inside 1 device.

On a decent high speed network (properly configured) an additional hop may add &lt;2ms to the overall latency. As a small example I ran a simple ping from one of my boxes on the UUNET backbone to the edge of the Sentech network:

ping 66.18.67.26
--- 66.18.67.26 ping statistics ---
17 packets transmitted, 17 received, 0% packet loss, time 16016ms
rtt min/avg/max/mdev = 7.137/9.959/11.206/1.002 ms, pipe 2

and this is travelling over 9 hops ... from uunet via jinx to sentech...


<hr noshade size="1"><font size="1"><div align="right">8:00 Thursday 26 August 2004 is <font color="red">D-Day</font id="red">....
<s> 16 17 18 19 20 <font color="green">21 22</font id="green"> 23 24</s> 25 <font color="red"><b>26</b></font id="red">
</div id="right"></font id="size1">

 

[ProAsm]

A rough guide on how the Towers are connected.

Radiokop is the main HUB and is where the main Server and all routers are housed.
Theoretically anyone connecting directly to Radiokop should have the best connection in the whole country as there are no other hops, also no links or backhauls.
Just off the top of my head, there are several "grid starters" and these could be like Brixton, Helderkruin, Sunnyside and a few others (think 7 in Gauteng)
Each one of these are connected to Radiokop via a 155 mbit microwave link.
Now each of these will have up to 8 Towers connected to them with 35 mbit microwave links.
Each City will also have a Hub, but is in actual fact another 'grid starter' like Brixton and is connected to Radiokop via Fibre (IS).
In some vary rare instances you may find a single Tower acting as a 'repeater' for a Tower placed further down the chain, although as I say this is rare but there are a few and they will share the 35 mbits.

So as you can see from the above that all the brains, routers and server sits at Radiokop and everything is controlled form there.
All towers act as "message passer oners" unless otherwise stated by the routers.
Finally there is a pipe from Radiokop to IS which is the peering point into the Internet.
Basically, "this side" of Radiokop is like a huge Intranet.

The data between the Tower and the server or routers at Radiokop all travel at max speed (3 mbits/sec) regardlesss of what package you got, 128, 256 or 512.
Currently from the Server out onto the Internet is where your package is capped to 128, 256 or 512.

As you can see there is no way a NodeB can slow you down or speed you up as its just a dumb message passer onner.

[]


<hr noshade size="1"><center><font color="blue">MyWireless Stuff</font id="blue"></center>

 

[regardtv]

Thanks for that ProASM ... so we've got multiple star configuration going ... I take it then that any tower numbered 82 would be a "grid starter" ....

/glares at Radiokop 1.2km away ... no damn node in my direction so I have to use Mintek 5.6km away ... aarrggh

<hr noshade size="1"><font size="1"><div align="right">8:00 Thursday 26 August 2004 is <font color="red">D-Day</font id="red">....
<s> 16 17 18 19 20 <font color="green">21 22</font id="green"> 23 24</s> 25 <font color="red"><b>26</b></font id="red">
</div id="right"></font id="size1">

 

[Wireless2]

ProAsm dont you mean 128, 24 or 512 ??

care to comment on whats going on with the 256K Packages ??

We are dying here

 

[aborg]

Thx ProAsm - that was quite informative, I must say though when my 256k package still had speed - it would be much faster from fourways then when I am at Radiokop where my mom stays.

 

[flyboy]

proasm,

I tend to disagree what you say about radiokop. I live a few metres from STP in radiokop. Downloading email is well below 128kbps, browsing speed also pathetic. Although I connect directly to the backbone which terminates at radiokop things dont seem any different speed wise!
tracert's are also pathetic (if u wanna see let me know)

 

[gripen]

Actually what he is saying is that while you are right at STP.. you can still get sucky speeds from upstream. The only benefit you have is less congestion on the path to STP (which in this case is nothing). You are still subjected to the same RBM (and hard capping) as everyone else.