Proxy block implemented - what's next

[regardtv]

Ok,

So Sentech read and reacted to TheRodents revelations.

I decided to use Rodent's "standard" test:
--21:48:42-- http://tucows.is.co.za/files4/DCPlusPlus-0.241.exe

21:57:08 <b>(3.96 KB/s)</b> - `DCPlusPlus-0.241.exe.1' saved [2044011/2044011]

Ok, so we've got one issue of of the way.... What is to be done next ?

Rather than just "monitoring" the excessive users - cut them back and see if it actually FIXES the problem ? If you know that a specific user is eating the bandwidth alive, cut them back to a single 128K package...

(And if you tell me that you cannot restrict people to single logins and enforce the bandwidth packages, I'm going to drop my head...and give up).

If not, come month end we're going to be in exactly the same position as the last time Sentech made a promise - NON-DELIVERY !

One can hope that someone from Sentech will "SINK SO LOW" as to make an ATTEMPT to contact me either via the forum or e-mail.

R

 

[TheRoDent]

The next thing to tackle is: Their microwave backlinks, and whatever bandwidth management tool they might be using.

It's clear from the tcptrace's from last weekend that there is quite a lot of unacceptable packet loss, which makes TCP freak out.

Whenever packet loss occurs, TCP believes it to be network congestion. When congestion occurs TCP decides that in order to reduce the network congestion, it has to back off, so the receiver starts ack'ing slower, and the sender starts transmitting slower until the packet loss has gone away.

I've mentioned this before in another thread.

All the TCP trace's we took this weekend show retransmissions (packet loss) of between 2-5%. This can be due to either:

1. A very ill designed bandwidth management tool, that looks at the sent average and starts throwing some packets away once it deems a specific connection to have gone above the "throttled" limit. Cheap bandwidth management tools do this, as well as Cisco's MPLS on shaped links.

2. It could be due to our lovely lossy IPWireless "air interface", and the signalling of these modems. The 4.2.1.8 firmware reduced the packet loss to near nothing, and then suddenly the packet loss came back somewhere in May? (Am I right there)

3. Network congestion due to the the cross-tower microwave links get congested, may cause packet loss.

I find it difficult to think that the problem is with 2) since I doubt IPW would have put this much time into a technology, that can go up to 3Mbps, but is so lossy that it limits actual thruput.

Seeing as Sentech announced a Microwave upgrade in CapeTown on their bulletin board, stating that this is the fix to all fixes for the problems down there, it may very well hold true for up here too. Except they're doing squat about it? Or are they.

My votes go for - most likely - option 3, and a good probable number 1, or a combination of the two.

A clever, and expensive bandwidth management tool will instead of discarding packets on connections that are going above the "set limit" start interfering with the speed at which TCP segments are acknowledged between the sender/receiver. It will also buffer the data so that packet loss doesn't occur (same as router queueing) until both have reduce their rate to an acceptable level. Of course, these things aren't cheap.

And we know Sentech are CHEAP, when it comes to spending on things that might make the service better.

<center><h6> MyWireless <s>Hacks</s> Tweaks & Tech Info || Have you checked the fawking FAQ? <br /> <font color="red">Tired of Sentech's bad service? Want to compare speeds? We at least listen...</font id="red"></h6></center>

 

[regardtv]

I think this topic should go sticky and be strictly moderatated to make it EASY for Sentech to notice ;-)

Nah, they'll notice it anyway - they have all admitted to reading here ;-)

My opinion as to things to look into:

1) Rodent's #1 --&gt; independent review of the current system and ENSURE that rate limiting is properly implemented
2) Rodent's #3, also mentioned by both of us before... review your tower interlink topology.
3) Push IPW towards v5 firmware... IPW and others have agreed that this "should" decrease packet loss even further
4) Improve customer feedback and involve them as a test base. We are paying beta testers after all, use us appropriately.

Just my 2c's worth.

I need to buy you a drink Rat ;-)

 

[TheRoDent]

Oh, and from testing, it appeard that those transparent cache have some sort of rate limit too. The max I could get for an uncached item from it was 5.25K/s (sound more or less familiar to 128k users?)

However, once it had been cached, I could get it at the top ADSL speed of 50Kb/s.

I believe that they may be using bandwidth pools on the transparents too.


<center><h6> MyWireless <s>Hacks</s> Tweaks & Tech Info || Have you checked the fawking FAQ? <br /> <font color="red">Tired of Sentech's bad service? Want to compare speeds? We at least listen...</font id="red"></h6></center>

 

[chuckl]

We start checking the rest of the network. Sntsick is obviously incapable in my opinion, but this forum has many capable people. Networking tools are available, find the bottlenecks.

To the barriers

 

[regardtv]

Are different packages served by different cache servers ?... Someone with a 512K please report in ;-)

I seem to keep getting assigned to stpxc02.sentechsa.net (168.210.90.181)

Same for everyone ?

I see only 2 proxies:
stpxc01
stpxc02

Interesting that one can pull the complete A record list from their services - not exactly best practices...3365 records extracted.

R

 

[TheRoDent]

I keep getting stpxc02 as well, on 128k

<center><h6> MyWireless <s>Hacks</s> Tweaks & Tech Info || Have you checked the fawking FAQ? <br /> <font color="red">Tired of Sentech's bad service? Want to compare speeds? We at least listen...</font id="red"></h6></center>

 

[lewstherin]

What about the infosat proxies? Rodent mentioned they were also left wide open. Have they been closed too now?

<font color="blue">Telkom needs a leash, ICASA needs some guts, and the </font id="blue"><font color="red">SA consumer</font id="red"><font color="blue"> needs to make it happen</font id="blue">

 

[DFantom]

How can I check what cache server I am getting?

I have been performing my own tests for over a week on the sentech network and found major DNS server related problems. Spoke to NoWire who confirmed there is major problems (they couldn't explain exactly what but the impression was if a bad DNS request is sent it can crash the DNS server). The DNS servers failing could account for slow downs in requests where host names are needed and no issues where IP's are used (such as a post I read a few days back where someone stated HTTP was slow but P2P wasn't).

--
DFantom
Sentech:Tower 82:22% Signal:512K Package

 

[Dean_Henstock]

Sentech uses a bandwidth managment tool developed by a guy who works for a company called tribe, it controls all there bandwidth for their VSAT and MW system, however I think they should get a packteer in, that should sort out the problems.

Keep Surfing