Crystal Web (NEW ISP) ADSL Feedback

[Crystal Web]

We've been monitoring the peering links between us and Diablo servers to test for issues, and it's mostly good news. The data shows minor downtime (not on our network) and good pings, in a fairly stable range, although this specific test still runs over the IPC network so there's no accounting for any jitter or other issues on the Telkom route. Here are the results since last week Friday:

In other words the connection is good and is stable, but the issue is packet loss, and this is because of the international issue. So expect far, far better performance on the gaming front very soon, in line with the sub-200ms range you see above. So far we know there is packet loss on the route and lag and jitter are not caused by routing issues (phew [although we want to have a stern word with ntt and telia nonetheless]), we know why, and there's a plan as well as contingency if the plan is delayed at all.

 

[Sinbad]

We've been monitoring the peering links between us and Diablo servers to test for issues, and it's mostly good news. The data shows minor downtime (not on our network) and good pings, in a fairly stable range, although this specific test still runs over the IPC network so there's no accounting for any jitter or other issues on the Telkom route. Here are the results since last week Friday:

In other words the connection is good and is stable, but the issue is packet loss, and this is because of the international issue. So expect far, far better performance on the gaming front very soon, in line with the sub-200ms range you see above. So far we know there is packet loss on the route and lag and hitter are not caused by routing issues (phew [although we want to have a stern word with ntt and telia nonetheless]), we know why, and there's a plan as well as contingency if the plan is delayed at all.

Sweet!
Any updates on the fix? Need another pair of hands?!

 

[Phinix]

@Crystal Web
In regards to the exchange trail program, how often will we receive data/updates on the exchange?
(Just curious)

 

[saa044]

Sweet!
Any updates on the fix? Need another pair of hands?!

They don't like sharing their donuts and coffee.

 

[Crystal Web]

@Crystal Web
In regards to the exchange trail program, how often will we receive data/updates on the exchange?
(Just curious)

We're pouring over the data constantly. I was going to start sending out regular data to trial customers tonight, but I found a problem with my algorithm that resulted in timing mismatches in the calculation that was supposed to exclude measurement latency from the equation, so I tweaked the algorithm again today to account for this. I then have to spend some time pouring over the data again when I have a chance to ascertain the accuracy of the information.

In a nutshell, I was concerned that sending the data may have been irresponsible of us, and we're now collecting, analysing, and tweaking calculations to give better and more granular data.

 

[Crystal Web]

They don't like sharing their donuts and coffee.

Hey, this ain't Bidorbuy.

 

[Crystal Web]

We're pouring over the data constantly. I was going to start sending out regular data to trial customers tonight, but I found a problem with my algorithm that resulted in timing mismatches in the calculation that was supposed to exclude measurement latency from the equation, so I tweaked the algorithm again today to account for this. I then have to spend some time pouring over the data again when I have a chance to ascertain the accuracy of the information.

In a nutshell, I was concerned that sending the data may have been irresponsible of us, and we're now collecting, analysing, and tweaking calculations to give better and more granular data.

Just to give you some idea of what this sort of thing requires - if you look at the above Diablo tests, we have 13,000 data points for analysis. The exchange/line checking system now has around 60,000 data points for the same time-frame that are analysed and compared against a second set of 60,000 data points, and then rebased. And that's just for 4 days, and for a single exchange.

So getting the calculation correct is paramount before sending out any information.

 

[ProfA]

Anybody else downloading NZBs using a server other than nextgennews? Like Sunnyusenet? If so, any problems?

 

[AfricanTech]

We're pouring over the data constantly. I was going to start sending out regular data to trial customers tonight, but I found a problem with my algorithm that resulted in timing mismatches in the calculation that was supposed to exclude measurement latency from the equation, so I tweaked the algorithm again today to account for this. I then have to spend some time pouring over the data again when I have a chance to ascertain the accuracy of the information.

In a nutshell, I was concerned that sending the data may have been irresponsible of us, and we're now collecting, analysing, and tweaking calculations to give better and more granular data.

pedantic mode on

poring

 

[Phinix]

We're pouring over the data constantly. I was going to start sending out regular data to trial customers tonight, but I found a problem with my algorithm that resulted in timing mismatches in the calculation that was supposed to exclude measurement latency from the equation, so I tweaked the algorithm again today to account for this. I then have to spend some time pouring over the data again when I have a chance to ascertain the accuracy of the information.

In a nutshell, I was concerned that sending the data may have been irresponsible of us, and we're now collecting, analysing, and tweaking calculations to give better and more granular data.

Just to give you some idea of what this sort of thing requires - if you look at the above Diablo tests, we have 13,000 data points for analysis. The exchange/line checking system now has around 60,000 data points for the same time-frame that are analysed and compared against a second set of 60,000 data points, and then rebased. And that's just for 4 days, and for a single exchange.

So getting the calculation correct is paramount before sending out any information.

I understand you completely. I would also want to be 100% sure that what you are sending has validity and is accurate. Take your time with it, as it is a luxury to have, not a immediate need.

I look forward to being able to dive into the data when it's ready.

 

[Paul Hjul]

pedantic mode on

poring

Its the pedants rule: We are being pedantic about one thing you'll make a mistke about another ...

(like forgetting to switch off pedantic mode)

[and yes there is a deliberate error]

 

[Crystal Web]

/pedantic mode on

poring

FTFY

 

[Replay]

This is no time to discuss infant ISPs...get your arses over to the WC soccer thread

YOU KNOW WHO YOU ARE!!

 

[Zenbaas]

It looks like CW has been working day and night to optimise their imported unit. One can see the improvements on an hourly basis!
By the looks of it, flying a Jumbo jet would have been easier.
I feel priviledged be be one of CW's first customers/ beta testers.

NZB being pulled down at 100kb/s at the moment on 10mb line.(edit down to 11kb/s now. Seems to be dying at the moment)

 

[HavocXphere]

...

Informal testing verdict...on stable line & solid exchange....CW vs TI. Testing the 108 address as that is the main traffic one if I'm eyeballing it right.

Both latencies rock stable...at 333 and 385 respectively (50ms diff!)...but occasional CW timeouts.

If D3 is TCP/IP based then those timeout might explain the spikes.

*These are ICMP tests which per Sinbad might not reflect the gaming reality 100%....though I'm inclined to think its a good approximating since afaik D3 has low throughput requirements

 

[Sinbad]

Informal testing verdict...on stable line & solid exchange....CW vs TI. Testing the 108 address as that is the main traffic one if I'm eyeballing it right.

Both latencies rock stable...at 333 and 385 respectively (50ms diff!)...but occasional CW timeouts.

If D3 is TCP/IP based then those timeout might explain the spikes.

*These are ICMP tests which per Sinbad might not reflect the gaming reality 100%....though I'm inclined to think its a good approximating since afaik D3 has low throughput requirements

It wants about 500kbps if I remember right.

333 and 385 are actually bad btw. ICMP RTT SHOULD be sub 200.
Yes the timeouts are causing severe havoc.

Also, for others who were talkng about it - my PS3 netflix is pulling 400KB/sec on a 720HD stream

 

[Gatsby]

It wants about 500kbps if I remember right.

333 and 385 are actually bad btw. ICMP RTT SHOULD be sub 200.
Yes the timeouts are causing severe havoc.

Also, for others who were talkng about it - my PS3 netflix is pulling 400KB/sec on a 720HD stream

Line Speed?

 

[Sinbad]

Line Speed?

10

 

[Necuno]

I'd rather bet on UDP.

Informal testing verdict...on stable line & solid exchange....CW vs TI. Testing the 108 address as that is the main traffic one if I'm eyeballing it right.

Both latencies rock stable...at 333 and 385 respectively (50ms diff!)...but occasional CW timeouts.

If D3 is TCP/IP based then those timeout might explain the spikes.

*These are ICMP tests which per Sinbad might not reflect the gaming reality 100%....though I'm inclined to think its a good approximating since afaik D3 has low throughput requirements

 

[Sinbad]

Strongly doubt it's UDP. It's not a video stream where loss of a few packets will cause a glitch in the video. Loss of a few packets will result in game state data being out of sync between client and server.