Re: FIBER CUT: Dallas to West Coast
On Thu, 30 August 2001, Patrick Greenwell wrote:
I don't know what the goal of "basement dual-homers" is. I think that is the other thread, which I haven't been keeping up with.
My point was simply that the "basement dual-homers" probably want the same thing as the "people that matter", which is disparate paths to the Internet.
Maybe Bill Gates' or C. Michael Armstrong's homes have dual-entrance cable vaults, but I suspect most basement dual-homers don't have disparate paths to the Internet nor or really looking for that level of reliability. How many people choose a second voice long distance company for their homes? So what do basement dual-homers really want? Most folks seem to be trying to protect themselves from the business practices of their provider, not necessarily the physical problems. Are we solving the wrong problem precisely?
Except for the small problem that there is no guarantee that DNS servers will honor TTLs, that this solution would necessitate dialing down TTLs to the lowest possible value to maximize effectiveness which would greatly increase DNS-related traffic, and flapping could be really ugly.
DNS in lieu of a routing protocol make me uncomfortable.
Necessity is the mother of invention. Problems are rarely solved before there is critical need to solve them. Should we be trying to use IP addresses as permanent identifies, or names? If it became a critical issue, would programmers start/best practice using the second, third, etc IP address returned in the DNS query? If a SYN fails, should the IP stack flush the DNS cache entry and check for a fresh IP address? I agree, using DNS in lieu of routing makes me uncomfortable. However, we also need to remember the reliable "service" we are creating is really at Layer 8. There may be a need for some smarter middle-ware (I hate that term) which spackles over some of the bumps and gaps. IP emulates a reliable network over an unreliable infrastructure. Do we also need an App/IP layerto emulate reliable services over an unreliable IP network? I would rather add a new floor on top, than trying to jack up the foundation and change the basement.
On 30 Aug 2001, Sean Donelan wrote:
On Thu, 30 August 2001, Patrick Greenwell wrote:
I don't know what the goal of "basement dual-homers" is. I think that is the other thread, which I haven't been keeping up with.
My point was simply that the "basement dual-homers" probably want the same thing as the "people that matter", which is disparate paths to the Internet.
Maybe Bill Gates' or C. Michael Armstrong's homes have dual-entrance cable vaults, but I suspect most basement dual-homers don't have disparate paths to the Internet nor or really looking for that level of reliability.
Sean, I think there is a disconnect in definitions. Randy's orignal rant defined anyone that didn't have >= /20 worth of space and 2xDS3's as a potential "basement dual-homer." By that definition there are a large number of "basement dual-homers" out there.
So what do basement dual-homers really want? Most folks seem to be trying to protect themselves from the business practices of their provider, not necessarily the physical problems. Are we solving the wrong problem precisely?
Well, as a "basement dual-homer"(by Randy's definition), I want disparate paths, as well as insulation from poor service and/or the complete failure(as a business) of a given organization. Having suffered at the hands of various bandwidth and large colo providers over the years("yes your service was down for 3 hours, where would you like your check for $27.12 sent?") I would never in good conscience recommend that an organization which wishes to provide 24/7 availability single-home, and based on my experience I reject the notion that there are magic providers out there that never go down.
Sean, I think there is a disconnect in definitions. Randy's orignal rant defined anyone that didn't have >= /20 worth of space and 2xDS3's as a potential "basement dual-homer."
Chuckle... And Randy just finished castigating an APNIC presenter who was making assumptions about >= (some number) and (small pipe size). His rant was that pipe size does not matter... 'cept when its in his interests to make a point. I 'spect. --bill
Sean, I think there is a disconnect in definitions. Randy's orignal rant defined anyone that didn't have >= /20 worth of space and 2xDS3's as a potential "basement dual-homer." Chuckle... And Randy just finished castigating an APNIC presenter who was making assumptions about >= (some number) and (small pipe size). His rant was that pipe size does not matter... 'cept when its in his interests to make a point. I 'spect.
did logic confuse you again. so sorry. randy
Sean, I think there is a disconnect in definitions. Randy's orignal rant defined anyone that didn't have >= /20 worth of space and 2xDS3's as a potential "basement dual-homer." Chuckle... And Randy just finished castigating an APNIC presenter who was making assumptions about >= (some number) and (small pipe size). His rant was that pipe size does not matter... 'cept when its in his interests to make a point. I 'spect.
did logic confuse you again. so sorry.
randy
ah, mr bush. being civil? --bill
On Thu, 30 Aug 2001, Patrick Greenwell wrote:
Sean, I think there is a disconnect in definitions. Randy's orignal rant defined anyone that didn't have >= /20 worth of space and 2xDS3's as a potential "basement dual-homer."
Hm, we have a /19 (although we only use a /21 at present, but E1 + ~ T1. If that means someone will be calling us basement dual homers, fine. As long as that doesn't mean we only get second rate connectivity, that is. I'm afraid the large networks are looking for an excuse to push the small ones out of business, by badmouthing "basement dual-homers" while in fact they themselves are the ones that bloat the global routing table.
By that definition there are a large number of "basement dual-homers" out there.
I don't think there are many "real" basement dual-homers. Getting a second line, an AS number and a decent router are too expensive if you don't make a reasonable amount of money on selling network services.
Well, as a "basement dual-homer"(by Randy's definition), I want disparate paths, as well as insulation from poor service and/or the complete failure(as a business) of a given organization. Having suffered at the hands of various bandwidth and large colo providers over the years("yes your service was down for 3 hours, where would you like your check for $27.12 sent?") I would never in good conscience recommend that an organization which wishes to provide 24/7 availability single-home, and based on my experience I reject the notion that there are magic providers out there that never go down.
My feeling is that the only way for a small network to compete with the large ones is by multihoming. If a small network connects to a larger network, it can only resell the larger network's services at a higher price and lower availability. So if the small network wants to compete on anything other than service, it has to connect to two large networks so its uptime is better than that of any of the those networks individually.
DNS in lieu of a routing protocol make me uncomfortable.
I agree, using DNS in lieu of routing makes me uncomfortable.
I would rather add a new floor on top, than trying to jack up the foundation and change the basement.
The basement is what's leaking. A new roof won't help. Also, DNS has loads of issues here which make it somewhat unsuitable for this purpose [1]. Another point, you would be bumping up the DNS cache sizes dramatically. Ass-u-me a 300 byte response for every query, and it really starts adding up, especially if you are talking about /24 or heaven forbid /32 entries. You would almost certainly run out of usable memory on your DNS caches long before you had solved the original problem. All told, better to fix the routing system than to lay it on DNS. [1] http://www.ehsco.com/misc/draft-hall-dns-data-00.txt which has not yet been submitted pending supplementary statistical research -- Eric A. Hall http://www.ehsco.com/ Internet Core Protocols http://www.oreilly.com/catalog/coreprot/
participants (7)
-
bmanning@vacation.karoshi.com
-
Eric A. Hall
-
Iljitsch van Beijnum
-
M. David Leonard
-
Patrick Greenwell
-
Randy Bush
-
Sean Donelan