Re: [nznog] Telecom's "peering" change of heart really isn't
Somebody wrote:
Oh, and on your point about universities, I think you will find that most (all?) of them already peer. UoA, which is far-and-away the largest university in the country, certainly peers at APE.
AUT doesn't peer with APE because it seems a little redundant as we have a connection through Orcon, who peer at APE We also connect to TCL and to KAREN.
Ian Batterbee wrote:
Somebody wrote:
Oh, and on your point about universities, I think you will find that most (all?) of them already peer. UoA, which is far-and-away the largest university in the country, certainly peers at APE.
AUT doesn't peer with APE because it seems a little redundant as we have a connection through Orcon, who peer at APE
You might want to ask Orcon to announce your /16 to APE peers then. A quick check from a couple of APE connected hosts shows traffic hitting TelstraClear, then AUT.
On 2/04/2007, at 9:20 AM, Alastair Johnson wrote:
Ian Batterbee wrote:
Somebody wrote:
Oh, and on your point about universities, I think you will find that most (all?) of them already peer. UoA, which is far-and-away the largest university in the country, certainly peers at APE.
AUT doesn't peer with APE because it seems a little redundant as we have a connection through Orcon, who peer at APE
You might want to ask Orcon to announce your /16 to APE peers then. A quick check from a couple of APE connected hosts shows traffic hitting TelstraClear, then AUT.
Assuming these providers also buy domestic transit from TelstraClear, AS_PATHs look like: - TelstraClear AUT - APE Orcon AUT Shortest AS_PATH wins. If they buy it from some other transit provider, who interconnects with TelstraClear: - <transit> TelstraClear AUT - APE Orcon AUT Assuming MED, etc. are the same, the oldest path will be selected. As APE routers get poked from time to time to add prefixes, the <transit> provider is likely to be the more stable. So overall, based on a few assumptions, TelstraClear wins. It's at this point that I encourage APE/WIX(1) peers to make sure you prefer routes learned over APE/WIX. LOCAL_PREF works well here to avoid AS_PATH length and age issues. -- Nathan Ward (1) And PNIX and SIX and CHIX and .. you get the idea.
Nathan Ward wrote:
Assuming these providers also buy domestic transit from TelstraClear, AS_PATHs look like: - TelstraClear AUT - APE Orcon AUT Shortest AS_PATH wins.
If they buy it from some other transit provider, who interconnects with TelstraClear: - <transit> TelstraClear AUT - APE Orcon AUT Assuming MED, etc. are the same, the oldest path will be selected. As APE routers get poked from time to time to add prefixes, the <transit> provider is likely to be the more stable.
Oldest path is configurable behavior - many people prefer the more deterministic lowest neighbor ID because it prevents neighbor triggered traffic switches. You're also assuming that all providers use the Route Servers and not bilateral peering; which is a risky assumption.
So overall, based on a few assumptions, TelstraClear wins. It's at this point that I encourage APE/WIX(1) peers to make sure you prefer routes learned over APE/WIX. LOCAL_PREF works well here to avoid AS_PATH length and age issues.
I'd generally assume anyone has a local_pref based on cost: ie: highest local_pref for customer (ie. "getting paid for") prefixes second highest for second lowest cost/efficiency (private peering) third highest for third lowest (public) fourth highest for transit (or other) Which fixes any problems with AS_PATH length. You raise some valid points... except the prefix in question wasn't announced by any peer that I could see in any case, so there was no determination being made.
On 2/04/2007, at 3:32 PM, Alastair Johnson wrote:
Oldest path is configurable behavior - many people prefer the more deterministic lowest neighbor ID because it prevents neighbor triggered traffic switches.
But it's not disabled default. My bet is that most people don't really consider the pros/cons of each in great detail, and instead opt for just-make-it-go.
You're also assuming that all providers use the Route Servers and not bilateral peering; which is a risky assumption.
From what I've seen, bi-lateral peering across the IXes is becoming less and less common.
I'd generally assume anyone has a local_pref based on cost: ie:
(speaking of risky assumptions.. :-)
highest local_pref for customer (ie. "getting paid for") prefixes second highest for second lowest cost/efficiency (private peering) third highest for third lowest (public) fourth highest for transit (or other)
Which fixes any problems with AS_PATH length.
Of course, I've seen a lot of networks that don't do things that you and I take for granted, though :-)
You raise some valid points... except the prefix in question wasn't announced by any peer that I could see in any case, so there was no determination being made.
Yep, the APE looking glass doesn't show any announcements either. Based on your assumptions, it appears that you're much more of an optimist than I. -- Nathan Ward
Nathan Ward wrote:
On 2/04/2007, at 3:32 PM, Alastair Johnson wrote:
Oldest path is configurable behavior - many people prefer the more deterministic lowest neighbor ID because it prevents neighbor triggered traffic switches.
But it's not disabled default. My bet is that most people don't really consider the pros/cons of each in great detail, and instead opt for just-make-it-go.
Perhaps. However the 'default' you refer to is also not common to all vendors, however. I would be interested to know how many people (ISPs) do elect to use "bgp bestpath compare-routerid" on their Cisco equipment. I'd imagine smaller/less experienced ASNs probably do leave it, but I suspect one or two ISPs may well have changed it.
You're also assuming that all providers use the Route Servers and not bilateral peering; which is a risky assumption.
From what I've seen, bi-lateral peering across the IXes is becoming less and less common.
Admittedly I haven't necessarily been near an IX recently, but I can't see bilateral going away fast - most people at APE seemed to prefer it when I had anything to do with configuring BGP. ihug's policy notwithstanding anyway.... There are IXs around that certainly enforce the behavior though, and it's one I have always wondered about for precisely this reason: the additional AS in the path introduces too many potential opportunities for the IX to be non-preferred in a default environment.... which is bad.
You raise some valid points... except the prefix in question wasn't announced by any peer that I could see in any case, so there was no determination being made.
Yep, the APE looking glass doesn't show any announcements either.
Which is one of the things I checked before posting to start with :P.
Based on your assumptions, it appears that you're much more of an optimist than I.
The glass is merely twice as big as it should be. aj. -- Am I really losing my negativity? This cannot go on.
On 1-Apr-2007, at 23:11, Nathan Ward wrote:
Assuming these providers also buy domestic transit from TelstraClear, AS_PATHs look like: - TelstraClear AUT - APE Orcon AUT Shortest AS_PATH wins.
I'd presume that highest LOCAL_PREF wins, actually. If I was running an ISP, I'd probably ensure that I had an import policy which promoted private peering > public peering > paid peering
transit in general.
Anybody who doesn't incorporate such policy into their network presumably doesn't care how much it costs (or how long it takes) to deliver a packet, and hence it doesn't matter which way the packets go. Assuming that any of this matters, customers of such an ISP will presumably migrate elsewhere naturally, and the problem (from the perspective of the AS originating the routes) will solve itself. Joe
Joe Abley wrote:
Anybody who doesn't incorporate such policy into their network presumably doesn't care how much it costs (or how long it takes) to deliver a packet, and hence it doesn't matter which way the packets go. Assuming that any of this matters, customers of such an ISP will presumably migrate elsewhere naturally, and the problem (from the perspective of the AS originating the routes) will solve itself.
What has been lacking thus far is any objective measurement of customer experience across different ISPs. Customers are not keen to move ISPs, change email addresses, change bank APs etc if they are not assured of a better service. If ISPs would adopt some neutral measurement capture, then there would be a way to say "Blah ISP is better because they <do thing X>, and Blerg ISP is worse because they don't" Some of the work Nevil Brownlee presented at the conference could feed into this. So could the NLANR AMP stuff (now many ISPs have this deployed again?). Even the nzdsl speed test stuff. I don't know if there is a speedtest server on the WIX/APE, but if there were then you'd be able to get a good appreciation of how the different ISPs fared when it came to a similar service. Dean
On Thu, April 5, 2007 08:57, Dean Pemberton wrote:
Joe Abley wrote:
Anybody who doesn't incorporate such policy into their network presumably doesn't care how much it costs (or how long it takes) to deliver a packet, and hence it doesn't matter which way the packets go. Assuming that any of this matters, customers of such an ISP will presumably migrate elsewhere naturally, and the problem (from the perspective of the AS originating the routes) will solve itself.
What has been lacking thus far is any objective measurement of customer experience across different ISPs.
the very nature of of the phrase "customer experience" is subjective...
Customers are not keen to move ISPs, change email addresses, change bank APs etc if they are not assured of a better service.
If ISPs would adopt some neutral measurement capture, then there would be a way to say "Blah ISP is better because they <do thing X>, and Blerg ISP is worse because they don't"
define "better" - to which destinations from which sources and for what cost?
Some of the work Nevil Brownlee presented at the conference could feed into this. So could the NLANR AMP stuff (now many ISPs have this deployed again?). Even the nzdsl speed test stuff.
I don't know if there is a speedtest server on the WIX/APE, but if there were then you'd be able to get a good appreciation of how the different ISPs fared when it came to a similar service.
a millisecond or two? a kilobit or three? are you testing with an o/s which has had network setting optimized? does the average user have those same optimized settings? would 90% of dsl users inside nz even notice the difference? if 95%+ of the content that people want is hosted offshore, you can peer at two or twenty-nine or 300 locations within new zealand and it isn't going to change the "customer experience". in fact, it would likely decrease performance as the overhead of maintaining more and more hardware and infrastructure increases and traffic engineering is significantly complicated. peering can be a good thing, but it isn't the panacea that it is often sold as within new zealand. /joshua -- A common mistake that people make when trying to design something completely foolproof is to underestimate the ingenuity of complete fools. - Douglas Adams -
Joshua, Customer experience is very subjective - but tends to be the basis by which companies try to attract new customers, and by which customers choose to leave their current provider. So objective, subjective or whatever, it needs to be addressed. Similar with 'Better'. You seem to be attacking the fact that these things are not nailed down to "Latency to site X is less than Y ms" or "There is under Z% packet loss to site X", when in fact thats not what the majority of customers care about anyway. I agree that 'peering' (whatever that means) isn't a silver bullet to any particular problem. I would suggest that we work harder to define the perceived problem, which I suspect is defined somewhere in the middle of: "X% packet loss and RTT of Yms to site Z" and "My Internet seems really slow today" Dean joshua sahala wrote:
On Thu, April 5, 2007 08:57, Dean Pemberton wrote:
Joe Abley wrote:
Anybody who doesn't incorporate such policy into their network presumably doesn't care how much it costs (or how long it takes) to deliver a packet, and hence it doesn't matter which way the packets go. Assuming that any of this matters, customers of such an ISP will presumably migrate elsewhere naturally, and the problem (from the perspective of the AS originating the routes) will solve itself.
What has been lacking thus far is any objective measurement of customer experience across different ISPs.
the very nature of of the phrase "customer experience" is subjective...
Customers are not keen to move ISPs, change email addresses, change bank APs etc if they are not assured of a better service.
If ISPs would adopt some neutral measurement capture, then there would be a way to say "Blah ISP is better because they <do thing X>, and Blerg ISP is worse because they don't"
define "better" - to which destinations from which sources and for what cost?
Some of the work Nevil Brownlee presented at the conference could feed into this. So could the NLANR AMP stuff (now many ISPs have this deployed again?). Even the nzdsl speed test stuff.
I don't know if there is a speedtest server on the WIX/APE, but if there were then you'd be able to get a good appreciation of how the different ISPs fared when it came to a similar service.
a millisecond or two? a kilobit or three? are you testing with an o/s which has had network setting optimized? does the average user have those same optimized settings?
would 90% of dsl users inside nz even notice the difference?
if 95%+ of the content that people want is hosted offshore, you can peer at two or twenty-nine or 300 locations within new zealand and it isn't going to change the "customer experience". in fact, it would likely decrease performance as the overhead of maintaining more and more hardware and infrastructure increases and traffic engineering is significantly complicated.
peering can be a good thing, but it isn't the panacea that it is often sold as within new zealand.
/joshua
Some of the work Nevil Brownlee presented at the conference could feed into this. So could the NLANR AMP stuff (now many ISPs have this deployed again?). Even the nzdsl speed test stuff.
I don't know if there is a speedtest server on the WIX/APE, but if there were then you'd be able to get a good appreciation of how the different ISPs fared when it came to a similar service.
I've over 2 million tests done from NZ DSL connections (For National Speeds), and overall there is no real difference on average in speeds per ISP (Nationally)(For DSL users) (even considering where the speedtest server is @ Orcon). Regarding the Speedtest.net results showing Airnet #1 and ICONZ #2, as the overall stats are taken from every test done (even on faster connections), So they will be skewed on ISP's who have more tests done on Higher Speed Connections (Wireless,Fibre etc). International Speed for DSL users per ISP is a different story, but I don't have enough results for any conclusion. Thanks
participants (7)
-
Alastair Johnson -
Craig Whitmore -
Dean Pemberton -
Ian Batterbee -
Joe Abley -
joshua sahala -
Nathan Ward