Afrihost Capped ADSL Feedback (MTN) - New beginning

[sgs]

Something is definitely not happy there... With the issues seeming to worsen during peak times it could either be congestion on the Exchange or a system on our side not handling the network management as well as it should.

This got me wondering and I felt like I needed to understand more about this. So when the power came back early last night, I installed smokeping on a VM and let it at it.

I monitored 4 different sites. They all behave differently at different times of the day. If it was the exchange, they would have all exhibited latency at the same time. So from those results, I can confidently say that the exchange here is not the main culprit - at least not in these graphs.

Let me share - starting with www.afrihost.com:


Not too bad - looking OK tonight.

Moving on to a site in Canada (rogers.com) for some international comparison:


Again, as expected - not too bad with a few spikes.

On to News24.com:


Okay, something is up here during business hours. But nevertheless, I left the best for last - The Google itself, as represented by the 8.8.8.8 DNS server:


15% packet loss max over the day and 9% right now? Wow.

I don't know who you peer with for the local Google cache servers, but something here is seriously wrong. Maybe someone with more knowledge about ISP peering can fill in a few of the blanks?

(As I sit here typing this now, I can see some similar behaviour kick in and all 4 traces are suddenly starting to exhibit similar spikes. So maybe there is some exchange congestion in there as well. But that does not explain the packet loss to Google's cache servers.)

 

[AfriGenie]

Hi Afripeeps

I am looking to swap around my package but my decision will be based on the whole double data deal. What exactly is going to happen after 30 June? Will double data be extended or will some other policy come into effect?

Evening!

At this point there has been no decision made quite yet, but I'm relatively sure we'll continue with Double Data on Capped ADSL accounts, or if not Double Data then at least something similar.

We'll definitely have an official announcement before the 25th of June.

 

[AfriGenie]

This got me wondering and I felt like I needed to understand more about this. So when the power came back early last night, I installed smokeping on a VM and let it at it.

I monitored 4 different sites. They all behave differently at different times of the day. If it was the exchange, they would have all exhibited latency at the same time. So from those results, I can confidently say that the exchange here is not the main culprit - at least not in these graphs.

Let me share - starting with www.afrihost.com:
View attachment 219622

Not too bad - looking OK tonight.

Moving on to a site in Canada (rogers.com) for some international comparison:
View attachment 219624

Again, as expected - not too bad with a few spikes.

On to News24.com:
View attachment 219626

Okay, something is up here during business hours. But nevertheless, I left the best for last - The Google itself, as represented by the 8.8.8.8 DNS server:
View attachment 219630

15% packet loss max over the day and 9% right now? Wow.

I don't know who you peer with for the local Google cache servers, but something here is seriously wrong. Maybe someone with more knowledge about ISP peering can fill in a few of the blanks?

(As I sit here typing this now, I can see some similar behaviour kick in and all 4 traces are suddenly starting to exhibit similar spikes. So maybe there is some exchange congestion in there as well. But that does not explain the packet loss to Google's cache servers.)

One thing to bear in mind, although we do have local Google Cache servers, and for the most part all Google-based services will resolve to these servers, if you ping or run a traceroute to the Google 8.8.8.8 DNS server it will almost a always resolve to the international serve r specifically Mountain View in the US.

You can see in this traceroute:
traceroute 8.8.8.8
traceroute to 8.8.8.8 (8.8.8.8), 30 hops max, 40 byte packets
1 gw.aserv.co.za (196.33.227.1) 1.097 ms 2.970 ms 3.201 ms
2 bryanston-gw2.aserv.co.za (196.33.227.8) 0.725 ms 0.705 ms 0.675 ms
3 csw1-b-jup-bry-vl2616.ip.isnet.net (196.34.134.253) 1.049 ms 1.543 ms 1.550 ms
4 core2a-bry-gi1-0-0.ip.isnet.net (168.209.218.2) 1.546 ms core1a-bry-gi0-0-0.ip.isnet.net (168.209.217.1) 1.547 ms core1a-bry-gi1-0-0.ip.isnet.net (168.209.218.1) 1.495 ms
5 168.209.100.205 (168.209.100.205) 1.976 ms 1.962 ms core1a-bry-ge1-0-1.isnet.net (168.209.100.237) 1.498 ms
6 196.26.0.130 (196.26.0.130) 1.443 ms 168.209.1.179 (168.209.1.179) 0.893 ms 0.858 ms
7 72.14.205.16 (72.14.205.16) 1.710 ms 1.437 ms 1.769 ms
8 72.14.239.35 (72.14.239.35) 1.731 ms 72.14.239.117 (72.14.239.117) 2.070 ms 2.064 ms
9 google-public-dns-a.google.com (8.8.8.8) 1.575 ms 1.574 ms 1.555 ms

This is from our Data Center in Gallo Manor, so it will be a bit better than from an ADSL connection.

Are you seeing your latency increase noticeably at the moment for local or international servers?

 

[Type RSA]

Hey guys, what are things like on the capped Afrihost accounts lately. Good enough to come back yet?

 

[DeathStrike]

For me no issues. But people are complaining about some issues.
Get 1gb account to test?

 

[AfriGuy]

For me no issues. But people are complaining about some issues.
Get 1gb account to test?

Best idea
I agree dude, always best to test things out first hand!

 

[Type RSA]

Best idea
I agree dude, always best to test things out first hand!

I have a 1gb free account. But with many years of doing this I have found that everything is always hunky dory on the 1gb account and then crap when you sign up. Sure none of their shaping etc rules apply to the 1gb free account.

 

[caroper]

Hey guys, what are things like on the capped Afrihost accounts lately. Good enough to come back yet?

If you don't want to use Chrome and Google Docs, Stream in the evenings in Cape Town or play any on-line games, then it is fine
My Telkom Shaped account out performs Afrifost Capped or Business Uncapped for streaming or working with Google Apps.

 

[Johnatan56]

If you don't want to use Chrome and Google Docs, Stream in the evenings in Cape Town or play any on-line games, then it is fine
My Telkom Shaped account out performs Afrifost Capped or Business Uncapped for streaming or working with Google Apps.

Online games in the evening* (past about 6.30pm). Before that everything is fine, after that unplayable (e.g. League of Legends everything just stops for up to 10 seconds and then it happens in super speed. Same for Heroes of the Storm, haven't tried Smite yet (writing exams and haven't really had time)).

Swapping to a WebAfrica account fixes all the issues.

 

[AfriGuy]

I have a 1gb free account. But with many years of doing this I have found that everything is always hunky dory on the 1gb account and then crap when you sign up. Sure none of their shaping etc rules apply to the 1gb free account.

Our 1GB free account is unshaped, as are our normal Capped and Business accounts.

 

[frobbertze2]

Just my 2 cents on whole Google services slowdown going on at Afrihost.

Currently google runs QUIC over udp on port 80 and 443, depending on what the user is accessing. This might be seen by the Afrihost shaper as traffic trying to bypass their shaping rules, as this is one way for torrents to also try and bypass shaping rules.
Now I know that normally business and capped packages are not being shaped, but still, some traffic must be getting priority on the network depending on what a user is using at the time,.
When the network does start to get busy and some protocols get priority, this might throw the google services into the same category as P2P applications which gets lower priority than other services, and is no longer seen as real time services.

Now I have no idea what software or hardware Afrihost or MTN uses to filter and prioritize traffic, so my explanation might be completely wrong.

 

[AfriGuy]

Just my 2 cents on whole Google services slowdown going on at Afrihost.

Currently google runs QUIC over udp on port 80 and 443, depending on what the user is accessing. This might be seen by the Afrihost shaper as traffic trying to bypass their shaping rules, as this is one way for torrents to also try and bypass shaping rules.
Now I know that normally business and capped packages are not being shaped, but still, some traffic must be getting priority on the network depending on what a user is using at the time,.
When the network does start to get busy and some protocols get priority, this might throw the google services into the same category as P2P applications which gets lower priority than other services, and is no longer seen as real time services.

Now I have no idea what software or hardware Afrihost or MTN uses to filter and prioritize traffic, so my explanation might be completely wrong.

That's the confusing bit, QUIC is there on accounts that are both shaped and unshaped. We basically have shaping pools setup, so Capped and Business run through one and Uncapped (when shaped) runs through another. Obviously this is a crude example, the actual process behind the shaping is much more elegant.

But what happens with unshaped accounts, the shaping systems sees the traffic (any traffic) and just lets it run. I believe we've even tried prioritising certain protocols before and even that didn't seem to work. So somewhere, something may be going wrong, or QUIC is a bit finicky. My personal experience, it works sometimes - others it doesn't.

Anyway - we've got guys looking into what we can do

 

[Mypro-D]

If you don't want to use Chrome and Google Docs, Stream in the evenings in Cape Town or play any on-line games, then it is fine
My Telkom Shaped account out performs Afrifost Capped or Business Uncapped for streaming or working with Google Apps.

 

[DeathStrike]

If you don't want to use Chrome and Google Docs, Stream in the evenings in Cape Town or play any on-line games, then it is fine
My Telkom Shaped account out performs Afrifost Capped or Business Uncapped for streaming or working with Google Apps.

I played games last night heroes of the storm and dota 2 worked fine for me.

During the day I did work with Google apps and ssh into servers based in EU and US without issues.

Also wife was streaming Netflix while I played games last night with no issues. 10mb line.

Based in cape town southern suburbs.

 

[Johnatan56]

Online games in the evening* (past about 6.30pm). Before that everything is fine, after that unplayable (e.g. League of Legends everything just stops for up to 10 seconds and then it happens in super speed. Same for Heroes of the Storm, haven't tried Smite yet (writing exams and haven't really had time)).

Swapping to a WebAfrica account fixes all the issues.

Correction. 7.30pm exactly.

 

[AfriGenie]

I played games last night heroes of the storm and dota 2 worked fine for me.

During the day I did work with Google apps and ssh into servers based in EU and US without issues.

Also wife was streaming Netflix while I played games last night with no issues. 10mb line.

Based in cape town southern suburbs.

That's awesome to hear!

The issues we've seen pop up don't appear to affect all of our clients, but for those affected it can be very frustrating to see your latencies spike!

We're definitely working on improving the consistency of our Client's experiences!

 

[sgs]

One thing to bear in mind, although we do have local Google Cache servers, and for the most part all Google-based services will resolve to these servers, if you ping or run a traceroute to the Google 8.8.8.8 DNS server it will almost a always resolve to the international serve r specifically Mountain View in the US.
...
9 google-public-dns-a.google.com (8.8.8.8) 1.575 ms 1.574 ms 1.555 ms

I need to immediately make it public that Afrihost have disproven the theory of relativity and have found a way for data to travel faster than light

Seriously? 1.5ms to California? Those pings aren't leaving your data center.

My Google stuff was shaky at around 19:00 tonight. Better now. Still major packet loss on icmp to 8.8.8.8 though. Not so much to other servers. Pattern very similar to last night.

 

[AfriMan]

I need to immediately make it public that Afrihost have disproven the theory of relativity and have found a way for data to travel faster than light

Seriously? 1.5ms to California? Those pings aren't leaving your data center.

My Google stuff was shaky at around 19:00 tonight. Better now. Still major packet loss on icmp to 8.8.8.8 though. Not so much to other servers. Pattern very similar to last night.

Latency was generally good in the South last night, so that's a little worrying. My reports show that we didn't see the usual spike at peak window, so demand was more or less under control for the whole night, which would lead me to believe that clients should have had a relatively good experience through the night

 

[bgotapen]

That was a while ago, I'm sure stacks has changed since Oct 2014
Drop me a PM with the deets if you'd like me to have anything followed up on for you though.

look it has gotten to a point where I realise this is in the eyes of afrihost not a problem. 2% packetloss just does not concern your staff... so I am warnign the people who actually use your service for gaming.

If you have the interest of your customers at heart you will make that packetloss top priority above all else as this might be the cause for many of the issues people are describing in this thread.

but hey I am just a customer that uses your service and has frist hand experience of how absolutely horrible it is to game with your netowrks 2% packetloss... for over a year....

 

[AfriGuy]

look it has gotten to a point where I realise this is in the eyes of afrihost not a problem. 2% packetloss just does not concern your staff... so I am warnign the people who actually use your service for gaming.

If you have the interest of your customers at heart you will make that packetloss top priority above all else as this might be the cause for many of the issues people are describing in this thread.

but hey I am just a customer that uses your service and has frist hand experience of how absolutely horrible it is to game with your netowrks 2% packetloss... for over a year....

If the experience isn't the best for any client we take it seriously, and that's exactly what we'll do here
I've got your PM, I'll check things out and get it escalated for you.