[Fwd: RE: [nznog] Changes to Telecom ADSL Network]
I personally am happy with this topic being discussed on list.. however being an anonymous looking sender is a bit off putting. Many of the topic on this list i dont have a problem with, but i do like to know is going on with regards to the internet in NZ. This is a topic i have been asked about by many people i know who arent gamers but have been experiencing many issues with the latency of their connections when nearing max bandwidth usage. I did not experience any problems with the previous traffic shaping with the dropping of packets, however i have experienced some very weird issues with the new packet shaping when nearing the bandwidth limits of my jetstart connection. On Mon, 2003-11-24 at 07:29, Craig Spiers wrote:
Why was this posted to NZNOG?
If you have gripes with telecom, take it up with telecom?
I personally don't have a problem with the changes telecom have made.. And I believe it hasn't been done on purpose, but a side effect from the changing of equipment vendors..
-----Original Message----- From: Anonymous User [mailto:ominona(a)emailaccount.com] Sent: Sunday, 23 November 2003 6:26 p.m. To: adsl(a)lists.unixathome.org; nznog(a)list.waikato.ac.nz Subject: [nznog] Changes to Telecom ADSL Network
An interesting topic for discussion:
Telecom is installing new equipment in its ADSL network, which handles traffic congestion in a slightly different way than previously. Overview Telecom is installing new equipment in its network, which handles traffic congestion in a slightly different way than previously.
QoS 101, which may help in understanding what change has occurred. On the whole, I have sympathy for every provider and vendor in the industry when it comes to Quality of Service issues. This is the same debate that comes up with any QoS implementation and will always affect the end-user in some way. When choosing queueing strategies, policing and shaping, you generally have to run a compromise, it's choosing which compromise to run. So in general terms, TCP hates packet loss, as it has to cut it's rate and slow-start again. This results in what's known as the saw-tooth effect and poor performance when there are only one or two TCP sessions in play. Shaping is usually better, as this causes TCP to "adjust" it's windowing to the best possible throughput and delay is not noticable to the end-user.
From the other side, UDP hates delay, primarily as the applications running through UDP are sensitive to delay. Voice, gaming and video are all better at skipping a few frames rather than having to wait for data to arrive.
One good solution is to use multiple queues, one with policing and strict priority for interactive/gaming traffic, and at least one with shaping for TCP traffic. Unfortunately although the core infrastructure is generally capable of supporting this now, we still have some time to go before it is commonly deployed in Internet scenarios. I stopped using Jetstart, and switched to Jetstream PURELY due to the impact this had on file-transfers, but the new 256kbps Jetstart with shaping would be perfect for my kind of usage. I would agree with Telecom that more people would be better off with some type of shaping rather than policing, and if it doesn't help you for gaming, then preventing congestion of the link by not doing simultaneous transfers should work. If you are getting "very weird" issues, they would be worth documenting and raising with your service provider to figure it out. I know this is grossly over-simplified, I am happy to talk offline if anyone wants more info, or thinks I am talking BS. Arron Scott Cisco NZ -----Original Message----- From: kyle carter [mailto:kyle_nznog(a)feet.net.nz] Sent: Sunday, 23 November 2003 8:23 p.m. To: nznog Subject: [Fwd: RE: [nznog] Changes to Telecom ADSL Network] I personally am happy with this topic being discussed on list.. however being an anonymous looking sender is a bit off putting. Many of the topic on this list i dont have a problem with, but i do like to know is going on with regards to the internet in NZ. This is a topic i have been asked about by many people i know who arent gamers but have been experiencing many issues with the latency of their connections when nearing max bandwidth usage. I did not experience any problems with the previous traffic shaping with the dropping of packets, however i have experienced some very weird issues with the new packet shaping when nearing the bandwidth limits of my jetstart connection. On Mon, 2003-11-24 at 07:29, Craig Spiers wrote:
Why was this posted to NZNOG?
If you have gripes with telecom, take it up with telecom?
I personally don't have a problem with the changes telecom have made..
And I believe it hasn't been done on purpose, but a side effect from the changing of equipment vendors..
-----Original Message----- From: Anonymous User [mailto:ominona(a)emailaccount.com] Sent: Sunday, 23 November 2003 6:26 p.m. To: adsl(a)lists.unixathome.org; nznog(a)list.waikato.ac.nz Subject: [nznog] Changes to Telecom ADSL Network
An interesting topic for discussion:
Telecom is installing new equipment in its ADSL network, which handles
traffic congestion in a slightly different way than previously. Overview Telecom is installing new equipment in its network, which handles traffic congestion in a slightly different way than previously.
_______________________________________________ NZNOG mailing list NZNOG(a)list.waikato.ac.nz http://list.waikato.ac.nz/mailman/listinfo/nznog
Quoting "Arron Scott (ascott)"
QoS 101, which may help in understanding what change has occurred. ... <snip> ... So in general terms, TCP hates packet loss, as it has to cut it's rate and slow-start again. This results in what's known as the saw-tooth effect and poor performance when there are only one or two TCP sessions in play. Shaping is usually better, as this causes TCP to "adjust" it's windowing to the best possible throughput and delay is not noticable to the end-user.
Can't let this go by - TCP only uses slow start (multiplicative increase) when a session is starting. The sawtooth effect is from TCP's congestion management, i.e. when it starts to loose packets it halves its congestion window (multiplicative decsrease), then increases it again one packet per round-trip time (additive increase). Using UDP is all very well, but UDP is not congestion-aware. Games writers would do better to use newer protocols like SCTP which are designed to signal congestion to the application, allowing the game writer to do something about it, such as slowing the sending rate. Overall though I agree with Aaron: multiple queues with properties suited to whatever applications you're running is a 'best' solution, otherwise just have lots of bandwidth so as to avoid congestion. CHeers, Nevil ----------------------------------------------------------------------- Nevil Brownlee Director, Technology Development Phone: +64 9 373 7599 x88941 ITSS, The University of Auckland FAX: +64 9 373 7021 Private Bag 92019, Auckland, New Zealand ------------------------------------------------- This mail sent through University of Auckland http://www.auckland.ac.nz/
On Mon, Nov 24, 2003 at 15:26 +1300, Arron Scott (ascott) wrote:
QoS 101, which may help in understanding what change has occurred.
On the whole, I have sympathy for every provider and vendor in the industry when it comes to Quality of Service issues. This is the same debate that comes up with any QoS implementation and will always affect the end-user in some way.
Why then should the operator choose?
When choosing queueing strategies, policing and shaping, you generally have to run a compromise, it's choosing which compromise to run.
Policing, at last, we leave the illusion of "QoS" behind and start speaking to the subject.
Shaping is usually better, as this causes TCP to "adjust" it's windowing to the best possible throughput and delay is not noticable to the end-user.
Unless the end-user, the customer, the one who pays, isn't interested in throughput... so again, why is the choice made by the operator?
From the other side, UDP hates delay, primarily as the applications running through UDP are sensitive to delay. Voice, gaming and video are all better at skipping a few frames rather than having to wait for data to arrive.
So this will interfere best with video and VoIP, now would there be a conflict of interest here for an incumbent telco who wants to sustain voice over copper and enter the video market? You decide.
I know this is grossly over-simplified, I am happy to talk offline if anyone wants more info, or thinks I am talking BS.
I think, once you drop the "QoS" pretence, and discuss "policing" there is no reasonable criticism that can be levelled at your explanation. "Why We Don't Need QOS: Trains, Cars, and Internet Quality of Service" ObURL: http://www.bricklin.com/qos.htm
Arron Scott Cisco NZ
Hamish. -- Poetry is the revelation of a feeling that the poet believes to be interior and personal but which the reader recognises as his own.
participants (4)
-
Arron Scott (ascott) -
Hamish MacEwan -
kyle carter -
Nevil Brownlee