I think Dave has the right idea here. Given the lousy overall network performance that I (and others) are often seeing for months from varieties of service providers, I think the service providers should be forced to provide rebates. I frequently have 10% packet losses to get from where I am to the Bay area (via New York). And my service provider (CERFnet) is telling me that their service provider (Sprint) is not even answering to their trouble reports. Forwarded message:
From nobody Thu Nov 2 04:40:21 1995 X-Authentication-Warning: upeksa.sdsc.edu: nobody owned process doing -bs Message-Id: <199511021201.AA05016@quark.isi.edu> To: Kate Lance <clance@sol.newcastle.edu.au> Cc: J.Crowcroft@cs.ucl.ac.uk, end2end-interest@ISI.EDU Subject: Re: links on the blink In-Reply-To: Your message of "Thu, 02 Nov 95 15:26:44 +1100." <199511020426.PAA22322@lily.newcastle.edu.au> Date: Thu, 02 Nov 95 11:55:17 +0000 From: Dave Mills <D.Mills@cs.ucl.ac.uk>
jon, Kate, Lack of specific performance data is surely disappointing; however, I used to work for COMSAT, so understand the corporate mentality of such dinosaurs. Let me suggest we mount an ongoing experiment in which representative links of deserving carriers are monitored, perhaps with pings launched from cron jobs, and the results submitted for publication in respected national media, like ISOC, ACM CCR, NY Times, etc. One of my (failed) missions at COMSAT was to persuade the lawyers to approve a tariff filing for a packet satellite service. Their stated opposition was based on an assumption that COMSAT would have to rebate charges for those packets not actually delivered to the destination gateway. Our course is clear. File requests for refund with the cognizant public utilities commission. The FCC would of course laugh; however, the reaction of the <state> PUCs would be most interesting, especially if it was pointed out that the rationale for granting the license in the first place was service to the public sector. Dave
On Fri, 3 Nov 1995, Hans-Werner Braun wrote:
I think Dave has the right idea here. Given the lousy overall network performance that I (and others) are often seeing for months from varieties of service providers, I think the service providers should be forced to provide rebates. I frequently have 10% packet losses to get from where I am to the Bay area (via New York). And my service provider (CERFnet) is telling me that their service provider (Sprint) is not even answering to their trouble reports.
Well, I think the problem is that providers lock people in 1 and 2 year contracts so if people get bad service they are stuck. That is why I only have month to month service, if people don't like there T1 then they can quit. I think we need more providers to do that, I know of a lot users that are on providers that want to switch, but have 8 months left on their contract. Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite 5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest ---------------------------------------------------------------------------
On Sat, 4 Nov 1995, Nathan Stratton wrote:
On Fri, 3 Nov 1995, Hans-Werner Braun wrote:
I think Dave has the right idea here. Given the lousy overall network performance that I (and others) are often seeing for months from varieties of service providers, I think the service providers should be forced to provide rebates. I frequently have 10% packet losses to get from where I am to the Bay area (via New York). And my service provider (CERFnet) is telling me that their service provider (Sprint) is not even answering to their trouble reports.
Well, I think the problem is that providers lock people in 1 and 2 year contracts so if people get bad service they are stuck. That is why I only have month to month service, if people don't like there T1 then they can quit. I think we need more providers to do that, I know of a lot users that are on providers that want to switch, but have 8 months left on their contract.
From this I would say it is a very different issue, if a provider has quality of service problems within the own network: e.g. a NOC not answering, sluggish links, packet loss above a normal tolerable level. I would say, that in such cases any contract can be terminated, and sued. But, of course a customer must be certain that the problems are caused within his
Some comments from my side, which are, however, not official comments of IDT Internet Services: Internet works the way that a provider can only guarantee a standard of quality within the perimeter of the provider's networks. E.g. guarantee a customer that he is a certain number of hops away to the meetpoints, or hand out a latency matrix between POPs, guarantee that the packet loss is under a certain margin, and of course, guarantee a certain percentage of uptime. There is no way to talk about end to end connectivity quality assurance. I have at all times at least one customer raging about how bad we are reachable, and that his partners on other networks can only get to us with such and such a delay/packet loss/unavailability. The understanding must be that a provider can ony control the own network. How traffic is routed outside is mostly uncontrollable, because most people do not run route servers yet that would take a policy from the radb (I have no special policy in btw, but will do this as soon as we run our route servers). provider's network. That's where most people are not sure, of course, even most 'consultants' have in reality no clue how to check something like this. Mike
Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite 5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest ---------------------------------------------------------------------------
---------------------------------------------------------- IDT Michael F. Nittmann --------- Senior Network Architect \ / (201) 928 1000 xt 500 ------- (201) 928 1888 FAX \ / mn@ios.com --- V IOS
I have heard this argument very many times (and used it myself, and sure understand it at a technical level). It is a very network-centric argument serving a specific service provider. As a customer I don't buy it. You guys need to take a customer centric approach and make the customers happy. You sell yourselves and get money as services providers for the *Internet* not your *local environment*. If you only sell XYZnet services, not problem, but please then do not advertise you provide Internet services just because you are marginally connected to the rest of the world with no clue about how to make the NANOG and global system work. In the end this will be a market driven by customers, not service providers. Customers will make the rules and will determine whether a service is good or lousy. Look for other examples. If my power outlets would regularly drop to 50 volts (or zero) I would get quite irritated. "Power" is sold on the open market between service providers. I would not accept an argument from a local service provider that my power is always dropping because some service provider 2,000 miles away is screwing up. I would consider that to be their problem to watch out for such things and to coordinate it right. Same with phones. I don't care what region you are in, but if I would call you, above the 99th percentile, as long as you are close to your phone and pick up, my call will get through and work without significant service degradations *despite* the fact that there are at least three service prodivers (local, long distance, local) involved. Why? Because power and phone companies have their shit together on that stuff, and coordinate and cooperate because they do understand they are all in the same boat. Of course, obviously it is not a fairly new and anarchic environment there, but has grown quite well into coordination. Are you guys up to it, or would you need regulation do it for you? Y'know, life can be easier if your parents tell you what to do, if you cannot get your act together yourself. Less stress, too. Less flexibility as well, perhaps, but methinks we have to choose some optimization function here, and buy off on the cost. In the end, this is not a sandbox for having a good time. People today *depend* on their network connection, and that it works is importenat to them. You *have* to go beyond just thinking locally.
On Sat, 4 Nov 1995, Nathan Stratton wrote:
On Fri, 3 Nov 1995, Hans-Werner Braun wrote:
I think Dave has the right idea here. Given the lousy overall network performance that I (and others) are often seeing for months from varieties of service providers, I think the service providers should be forced to provide rebates. I frequently have 10% packet losses to get from where I am to the Bay area (via New York). And my service provider (CERFnet) is telling me that their service provider (Sprint) is not even answering to their trouble reports.
Well, I think the problem is that providers lock people in 1 and 2 year contracts so if people get bad service they are stuck. That is why I only have month to month service, if people don't like there T1 then they can quit. I think we need more providers to do that, I know of a lot users that are on providers that want to switch, but have 8 months left on their contract.
Some comments from my side, which are, however, not official comments of IDT Internet Services:
Internet works the way that a provider can only guarantee a standard of quality within the perimeter of the provider's networks. E.g. guarantee a customer that he is a certain number of hops away to the meetpoints, or hand out a latency matrix between POPs, guarantee that the packet loss is under a certain margin, and of course, guarantee a certain percentage of uptime.
There is no way to talk about end to end connectivity quality assurance. I have at all times at least one customer raging about how bad we are reachable, and that his partners on other networks can only get to us with such and such a delay/packet loss/unavailability.
The understanding must be that a provider can ony control the own network. How traffic is routed outside is mostly uncontrollable, because most people do not run route servers yet that would take a policy from the radb (I have no special policy in btw, but will do this as soon as we run our route servers).
From this I would say it is a very different issue, if a provider has quality of service problems within the own network: e.g. a NOC not answering, sluggish links, packet loss above a normal tolerable level. I would say, that in such cases any contract can be terminated, and sued. But, of course a customer must be certain that the problems are caused within his provider's network. That's where most people are not sure, of course, even most 'consultants' have in reality no clue how to check something like this.
Mike
Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite 5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest ---------------------------------------------------------------------------
---------------------------------------------------------- IDT Michael F. Nittmann --------- Senior Network Architect \ / (201) 928 1000 xt 500 ------- (201) 928 1888 FAX \ / mn@ios.com --- V IOS
Hans Werner - presumably frustrated about the quality of his internet service writes: Are you guys up to it, or would you need regulation do it for you? Y'know, life can be easier if your parents tell you what to do, if you cannot get your act together yourself. Less stress, too. Less flexibility as well, perhaps, but methinks we have to choose some optimization function here, and buy off on the cost. In the end, this is not a sandbox for having a good time. People today *depend* on their network connection, and that it works is important to them. You *have* to go beyond just thinking locally. Very relevant questions. Let me try to nudge the discussion into a slightly different direction. When people started talking to me earlier in the summer about internet business models, these issues were the number one argument. The chain of reasoning went something like: you can't have a mature internet business until you have guaranteed quality of service metrics. You can't have guaranteed quality of service metrics until you have settlements. It seems to me that Hans Werner's complaint above fits squarely into this mold of thought. If I am wrong I am sure he'll tell me. Could we talk a bit about where these conclusions may lead us? Settlements, as I understand them, might be traffic based or route based. If I use more of someone else's resources than he uses of mine, I owe him a payment. Perhaps the payment might be for gigabytes delivered or routes advertised. The payment process then would cascade from the largest down to the smallest providers as those lower in the food chain extracted money from those beneath them to pay the demands of those above them. The outcome of such a process would be significant. Traffic based settlemnts would add substantial accounting costs to doing business. Route charging I'd guess would add less. Again however would not the billing components of this play into the hands of the RBOCS and five largest IXCs who have the mainframes and software systems to handle it most easily? But if the big boys started it, how far would it or could it go before shattering the internet? I predict on the basis of what regional ISPs who have more dial up than leased line accounts tell me that they would disconnect from the big boys rather than accept the strangle hold of settlements. The result - viola - a fractured, balkanized internet. How are you going to guarantee the standards of service? PUC certification in each state? Just what the RBOCs would love. Everyone has to open a 7 by 24 NOC or you can't be in the business? Cisco 7000 routers become minimum gear? I know. Everyone has to meet standards set by the CIX in order to play. Seriously though - how are you going to do it? What will the rules be and how will they be made? Lets assume that this were done in the most benign possible way. The only way smaller isps could stay in business would be to pass the increased costs on to their customers. And then your friendly South dakota Internet service, would be like the nationals - $20 a month for 15 hours and $2 and hour for each additional hour. Not to mention the fact that the costs of entry would be catapulted so high as to exclude new start ups lacking sufficient outside financing. As Compu$erve remarked in July, the cost model of $20 a month for unlimited service is a suicidal one for this industry. Yeah undoubtedly if you have the overhead costs of an Hr Block or RBOC to satisfy. Let me frame the question starkly. To get the reliability that Hans Werner wants to what extent must we consolidate the industry and raise prices? Killing in the process what makes the internet a magnet of attraction for some millions of users? **OR** is there any alternative of letting a few build an inustrial strength network for those who demand it, an allowing a less costly more fault tolerant network to survive? If so how might this be done? Comments from policy makers for the big five would be welcome. ******************************************************************** Gordon Cook, Editor & Publisher Subscript.: Individ-ascii $85 The COOK Report on Internet Non Profit. $150 431 Greenway Ave, Ewing, NJ 08618 Small Corp & Gov't $200 (609) 882-2572 Corporate $350 Internet: cook@cookreport.com Corporate. Site Lic $650 Newly expanded COOK Report Web Pages http://pobox.com/cook/ ********************************************************************
On Sat, 4 Nov 1995, Hans-Werner Braun wrote:
I have heard this argument very many times (and used it myself, and sure understand it at a technical level). It is a very network-centric argument serving a specific service provider. As a customer I don't buy it. You guys need to take a customer centric approach and make the customers happy. You sell yourselves and get money as services providers for the *Internet* not your *local environment*. If you only sell
no, we sell internet access. Your argument would require carmakers to provide for sufficiently scenic driving environment ;-)
XYZnet services, not problem, but please then do not advertise you provide Internet services just because you are marginally connected to
we provide internet access services. Same with the phone: nobody guarantees you that you get through to china, isn't it?
the rest of the world with no clue about how to make the NANOG and global system work.
In the end this will be a market driven by customers, not service providers. Customers will make the rules and will determine whether a service is good or lousy. Look for other examples. If my power outlets would regularly drop to 50 volts (or zero) I would get quite irritated.
this is not a good comparison: we talk reachability here, analogy would be phone networks. Power: there is no geographical difference: 110Volts from Kansas look exactly the same as 110 Volts from Wisconsin. "Power" is sold on the open market between service providers. I would
not accept an argument from a local service provider that my power is always dropping because some service provider 2,000 miles away is screwing up. I would consider that to be their problem to watch out for
the service agreement here is not to reach that location, but to have power. and, btw: you do not get compensated for outages at all: if I must throwh away my freezer contents, they do not pay me for it. That's exactly like Internet access: if you cannot reach MIT because their link is down, we won't pay you your money back, and won't come up for any damages because you could not deliver a document or else.
such things and to coordinate it right. Same with phones. I don't care what region you are in, but if I would call you, above the 99th percentile, as long as you are close to your phone and pick up, my call will get through and work without significant service degradations *despite* the fact that there are at least three service prodivers
wrong: this week I triet to reach Chicago several times and got "sorry, your call cannot be completed at this time ...... try later". Can I now go to the phone company and say: "I did not reach my business partner to stop a deal and lost money, compensate me?".. I won't even try.
(local, long distance, local) involved. Why? Because power and phone companies have their shit together on that stuff, and coordinate and
two weeks ago one of these nice transformers on the poles exploded about 1.5 miles from here, sending a loud boom, and a fantastic power surge that burnt out a light bulb (my surge protector held). Don't even ask them to pay your stereo if it fries in such an incident.
cooperate because they do understand they are all in the same boat. Of course, obviously it is not a fairly new and anarchic environment there, but has grown quite well into coordination. Are you guys up to it, or would you need regulation do it for you? Y'know, life can be
Regulation would make it better:I could sit back and point to a regulation.
easier if your parents tell you what to do, if you cannot get your act together yourself. Less stress, too. Less flexibility as well, perhaps, ^^^^^^^^^^^^^^^^
see two lines above re regulations ;-)
but methinks we have to choose some optimization function here, and buy off on the cost.
In the end, this is not a sandbox for having a good time. People today
HOwever, I don't hate my job that much...
*depend* on their network connection, and that it works is importenat to them. You *have* to go beyond just thinking locally.
We do: we offer redundant dual homing and configure even your ospf or BGP for you. Even if you chose a different provider.
On Sat, 4 Nov 1995, Nathan Stratton wrote:
On Fri, 3 Nov 1995, Hans-Werner Braun wrote:
I think Dave has the right idea here. Given the lousy overall network performance that I (and others) are often seeing for months from varieties of service providers, I think the service providers should be forced to provide rebates. I frequently have 10% packet losses to get from where I am to the Bay area (via New York). And my service provider (CERFnet) is telling me that their service provider (Sprint) is not even answering to their trouble reports.
Well, I think the problem is that providers lock people in 1 and 2 year contracts so if people get bad service they are stuck. That is why I only have month to month service, if people don't like there T1 then they can quit. I think we need more providers to do that, I know of a lot users that are on providers that want to switch, but have 8 months left on their contract.
Some comments from my side, which are, however, not official comments of IDT Internet Services:
Internet works the way that a provider can only guarantee a standard of quality within the perimeter of the provider's networks. E.g. guarantee a customer that he is a certain number of hops away to the meetpoints, or hand out a latency matrix between POPs, guarantee that the packet loss is under a certain margin, and of course, guarantee a certain percentage of uptime.
There is no way to talk about end to end connectivity quality assurance. I have at all times at least one customer raging about how bad we are reachable, and that his partners on other networks can only get to us with such and such a delay/packet loss/unavailability.
The understanding must be that a provider can ony control the own network. How traffic is routed outside is mostly uncontrollable, because most people do not run route servers yet that would take a policy from the radb (I have no special policy in btw, but will do this as soon as we run our route servers).
From this I would say it is a very different issue, if a provider has quality of service problems within the own network: e.g. a NOC not answering, sluggish links, packet loss above a normal tolerable level. I would say, that in such cases any contract can be terminated, and sued. But, of course a customer must be certain that the problems are caused within his provider's network. That's where most people are not sure, of course, even most 'consultants' have in reality no clue how to check something like this.
Mike
Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite 5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest ---------------------------------------------------------------------------
---------------------------------------------------------- IDT Michael F. Nittmann --------- Senior Network Architect \ / (201) 928 1000 xt 500 ------- (201) 928 1888 FAX \ / mn@ios.com --- V IOS
---------------------------------------------------------- IDT Michael F. Nittmann --------- Senior Network Architect \ / (201) 928 1000 xt 500 ------- (201) 928 1888 FAX \ / mn@ios.com --- V IOS
I will not go into a point by point rebuttal here, even though I generally do not subscribe to your arguments. I am not planning on "winning" here, I just want to get the issues on the table and evaluate the solution space. Just let me ask you, as a customer who fairly frequently experiences 10% packet loss between major Internet locations across major service providers (no mom and pop shops in the middle or at the end points), how would you suggest I deal with that? My complaints to the involved service providers have typically gotten unanswered by the national service provider, and saying "we can't do anything about it except letting our national service provider know" by my regional service provider. There is no quality control at the inter-ISP level. I want to see that fixed. I don't care how, I do know that the current situation is intolerable. I believe that this is prime NANOG (and IEPG) business. NSF and the feds are out, with the NSFNET backbone dismantling, and the kitchen you asked for to cook (ahem!) in is all yours to get your tailfeathers burned in all by yourself.
I second hwb's observation that ISPs must take more responsibility for the overall QOS of the Internet infrastructure. IMHO, building business profits from providing IP services and not funding the development of routing protocols that provide Policy Based Routing and QOS hooks does not make good long term business. If end users, for example, could choose how their datagrams travel based on reliability and routing policy, then end-users would benefit greatly by enhancced IP services and all providers would benefit. The current, post NSF funding, days appears to resemble this paradigm: IP service providers take advantage of work by the NSF, IETF and code developers to create a commercial business. IP service providers (not all, of course, but most) make good profits and do not help fund (again not all, but most) the development of enhanced features in the IP development saga. Just sending packets "up stream" and then telling the end-user they do have any responsibility with "up stream networks" does nothing to improve the reliability of the Internet nor does it enhance the product offering. NANOG would appear to be an excellent vehicle for developing a strategic plan, as well as a funding strategy, to improve, enhance, and forward the development of these services. The IEFT is, for the most part, a volunteer organization. Where does all that CIX money go, BTW? Regards, Tim -- +--------------------------------------------------------------------------+ | Tim Bass | #include<campfire.h> | | Principal Network Systems Engineer | for(beer=100;beer>1;beer++){ | | The Silk Road Group, Ltd. | take_one_down(); | | | pass_it_around(); | | http://www.silkroad.com/ | } | | | back_to_work(); /*never reached */ | +--------------------------------------------------------------------------+
On Sat, 4 Nov 1995, Hans-Werner Braun wrote:
I will not go into a point by point rebuttal here, even though I generally do not subscribe to your arguments. I am not planning on "winning" here, I just want to get the issues on the table and evaluate the solution space. Just let me ask you, as a customer who fairly frequently experiences 10% packet loss between major Internet locations across major service providers (no mom and pop shops in the middle or at the end points), how would you suggest I deal with that?
Uh... Ignore it? 10% packet loss is quite within the normal range of parameters for a packet switching network such as the Internet. If you want 0% packet loss, you can lease your own private point-to-point lines.
my regional service provider. There is no quality control at the inter-ISP level. I want to see that fixed.
But nothing is broken. There is no inter-ISP level. ISP's buy access to the global network from and NSP and resell those access rights to you. Instead of millions of inter-ISP relationships, there is only one (or a few for multi-homed ISP's) relationship to negotiate and to manage.
I don't care how, I do know that the current situation is intolerable.
Well why didn't you say so in the first place! We thought you *DID* care how. Since you don't care how the problem is solved you will be happy to know that SPRINT and MCI et al. will be pleased to provide you with the performance guarantees that you require. Just get out your checkbook and call them on Monday morning. Make sure you tell them that price is no object.
I believe that this is prime NANOG (and IEPG) business. NSF and the feds are out, with the NSFNET backbone dismantling,
The NSF backbone is long gone (5 months or so), and the national backbones (note plural) at the core of the Internet are fast evolving into international webworks of fibre. As the networks grow and the infrastructure is deployed there are lots of pains. Live with them. In 10 years it will be over. Think of it like putting up with construction in your house and mud where the front yard should be. Once the construction is finished you will soon forget about it as you enjoy your new home. Michael Dillon Voice: +1-604-546-8022 Memra Software Inc. Fax: +1-604-542-4130 http://www.memra.com E-mail: michael@memra.com
Date: Sat, 4 Nov 1995 12:48:36 -0800 (PST) From: Michael Dillon <michael@memra.com> Subject: Re: links on the blink (fwd)
Uh... Ignore it? 10% packet loss is quite within the normal range of parameters for a packet switching network such as the Internet.
I must disagree here. 10% packet loss within national backbones is a problem to be fixed as soon as possible. It is not something to be tolerated. An examination of some of the interconnect points will find providers talking across media that is far past saturation, and is at capacity. The good news is that from my point of view, these things are being addressed. Not as quickly as everyone would like (including me), but it's happening.
my regional service provider. There is no quality control at the inter-ISP level. I want to see that fixed.
But nothing is broken. There is no inter-ISP level. ISP's buy access to the global network from and NSP and resell those access rights to you. Instead of millions of inter-ISP relationships, there is only one (or a few for multi-homed ISP's) relationship to negotiate and to manage.
Assuming Hans-Werner meant inter-NSP, I must differ with both of you.. There *is* quality control at the inter-NSP level. But there are things broken, as I said. There are also a bunch of folks working night and day to make sure it works as well as it does, and a bunch more trying to make it get better. Speaking only for myself... RobS
On Sat, 4 Nov 1995, Rob Skrobola wrote:
From: Michael Dillon <michael@memra.com>
Uh... Ignore it? 10% packet loss is quite within the normal range of parameters for a packet switching network such as the Internet.
I must disagree here. 10% packet loss within national backbones is a problem to be fixed as soon as possible.
Continuous 10% packet loss does need to be fixed. But the nature of Internet traffic is such that it can *NOT* be eliminated. There will always be bursts of packet loss like this no matter what the architecture. This may change if and when the nature of Internet traffic changes but for now that is a given.
It is not something to be tolerated. An examination of some of the interconnect points will find providers talking across media that is far past saturation, and is at capacity.
The good news is that from my point of view, these things are being addressed. Not as quickly as everyone would like (including me), but it's happening.
In fact, it was on the NANOG list here that Sean and somebody else recently discussed Sprint's and MCI's plans to add many two-way interchange points between their networks to take some load off the NAPS because the NAP architecture just wasn't working out in practice. In other words, the problem is known, has been publicly acknowledged, a solution has been discussed and NSP's have publicly announced that they are deploying that solution. Seems fine to me.
my regional service provider. There is no quality control at the inter-ISP level. I want to see that fixed.
But nothing is broken. There is no inter-ISP level. ISP's buy access to
Assuming Hans-Werner meant inter-NSP, I must differ with both of you.. There *is* quality control at the inter-NSP level.
Correct terminology means everything doesn't it? ;-)
things broken, as I said. There are also a bunch of folks working night and day to make sure it works as well as it does, and a bunch more trying to make it get better.
And some of us do appreciate your hard work and do understand that you don't have any magic wands to wave. Michael Dillon Voice: +1-604-546-8022 Memra Software Inc. Fax: +1-604-542-4130 http://www.memra.com E-mail: michael@memra.com
===== Michael Dillon previously wrote: ====
Uh... Ignore it? 10% packet loss is quite within the normal range of parameters for a
Well, Michael, considering TCP's sliding window and retransmission, I've seen 10% constant packet loss means you can get to the other side, but with such a pain and incredible patience, and certainly, is not tolerable. Jun -- o o o o o o . . . ____======_T__ ___=========_T__ ___==========_T__ o _____ || John J. Wu | | Sprint Intl. | | smail:john.wu | .][__n_n_|DD[ ====____ | ~{Nb>|~} | | (703)6895325 | | jun@hq.si.net | |
(________|__|_[________]_|____________|_|______________|_|_______________|_| __/oo OOOOO oo` ooo ooo 'o^o o^o` 'o^o o^o` 'o^o o^o`
Mike, Devious would it be for me to monkey in this debate; however, I'm sure HWB will pick up on these points. Ten percent loss is most definitely not within the envelope of acceptable service. The protocol models I have participated in developing and evaluating perform best at loss rates not exceeding one percent on average. My access is now from London, where the transatlantic loss rates are more like 30 percent. I would recommend a sentence to any advocate/employee/consultant making a claim such as yours to live with these rates for any length of time. Second, you suggest patience and that the problems should go away in ten years. Excuse me, I heard (and even said myself as NSF advisor and researcher) the identical words ten years ago. From my standpoint, things were approaching nominally good service five years agoa, but today things are degrading to the bad old days. I must conclude that the cycle over the next ten years is most likely to repeat. The Internet may get lots bigger and more ubiquitous, but I bet the discussion we are having now will be replayed verbatim then. Dave
The current method should probably be: locate where the packet loss occurs, forward it to your provider, your provider forwards it to his, and so on. Yes, there is no formal mechanism to my knowledge, but should we really make one? If we start to formalize then we have later to formalize the meetpoint relationships (who is responsible for packet loss on a meetpoint network?). A better point of pressure: chose a provider with shorter internal paths. On the Internet anyone can kick the tires of any network by sending probing data accross it. You get latency, hop count, and throughput of the smallest link in your path. The job to find this out is a consultant's job. The consultant knows about the Internet topology, and how to quickly find out if a provider will be ok for the wanted connectivity, or not. Scopes of connectivity can be global reachability (trading off maybe hopcounts and comparing on offered routes and latency), short paths (for virtual private networks or national connectivity), points of presence in your business locations. If everyone would chose, then the now big networks, that offer an impressive display of capital investment (called excessive hopcount by some people with negative attitude ;-) ), will probably redisign their routing (or why do some connections backtrack via both coasts and then to the next metropole 200 miles away?). I would say, that market pressure is the best regulatory agent: pinches directly into the providers pocketbook. I accept this and design for customers, and have customer needs and quality issues in mind when the network grows (which it does currently tremendously). Instead of organizational reglementation, I would put the current path of Internet routing forward: with current tools it is possible to include routing decisions of networks around a provider in that providers routing topology. It is just a matter of time to implement it. Mike On Sat, 4 Nov 1995, Hans-Werner Braun wrote:
I will not go into a point by point rebuttal here, even though I generally do not subscribe to your arguments. I am not planning on "winning" here, I just want to get the issues on the table and evaluate the solution space. Just let me ask you, as a customer who fairly frequently experiences 10% packet loss between major Internet locations across major service providers (no mom and pop shops in the middle or at the end points), how would you suggest I deal with that? My complaints to the involved service providers have typically gotten unanswered by the national service provider, and saying "we can't do anything about it except letting our national service provider know" by my regional service provider. There is no quality control at the inter-ISP level. I want to see that fixed. I don't care how, I do know that the current situation is intolerable. I believe that this is prime NANOG (and IEPG) business. NSF and the feds are out, with the NSFNET backbone dismantling, and the kitchen you asked for to cook (ahem!) in is all yours to get your tailfeathers burned in all by yourself.
---------------------------------------------------------- IDT Michael F. Nittmann --------- Senior Network Architect \ / (201) 928 1000 xt 500 ------- (201) 928 1888 FAX \ / mn@ios.com --- V IOS
HW, Eureka! I have confirmed the Astronautical Long Delayed Echo (ALDE) hypothesis and, by implication, confirmed the speed of light. The discussion in the now defunct NSF Network Technical Advisory Group (NTAG) circa 1985 with respect to ARPAnet and NSF service has returned after reflection due impedance mismatch from Alpha Centauri. The names have been altered in transit, but the time delay is accurate and the other words are near verbatim. Dave
power. and, btw: you do not get compensated for outages at all: if I must throwh away my freezer contents, they do not pay me for it. That's exactly like Internet access: if you cannot reach MIT because their link
If you are willing to pay for it, you can get what you want regarding power. There are companies that will gladly sell you dual diesel generators and companies that will sell you insurance in case the main power and both generators fail. If there are enough people who want and are willing to pay for < .001% packet loss, providers can and eventually will provide it. Good measurement systems are a necessary first step. Priority based packet routing will also help.
There is no way to talk about end to end connectivity quality assurance. The understanding must be that a provider can ony control the own network. How traffic is routed outside is mostly uncontrollable, because
You say you can control your own network, but I suspect much of it is based on telco supplied lines where you don't directly control the reliability. This is why companies have contracts. Within certain parameters (and for a price), they allow you to control the actions of other companies (and their networks).
Mike, Well, when it comes to evidence, traceroute does a pretty good job of identifying probable loss points. Perhaps I shouldn't reveal this tool, since it may result in action to disable the necessary router features. Meanwhile, I'm sure that HWB will confirm numerous instances where causual NTP servers have detected subtle problems on the links between them and that were undetected by any other means, including SNMP. Therefore, I can detect with some degree of confidence that some j-random provider is losing x percent of my packets between the y and z demarcations of their network. Dave
Good point. If there is no service model and penalties for bad performance, at least there should be a possibility for short time horizons. I am really curious how service providers believe that, say, 12-24 months from now customers will still just put up with the current situation. If I were them, I would plan for survival in that time frame and get the right things under way *now*, and before someone else changes things instead.
On Fri, 3 Nov 1995, Hans-Werner Braun wrote:
I think Dave has the right idea here. Given the lousy overall network performance that I (and others) are often seeing for months from varieties of service providers, I think the service providers should be forced to provide rebates. I frequently have 10% packet losses to get from where I am to the Bay area (via New York). And my service provider (CERFnet) is telling me that their service provider (Sprint) is not even answering to their trouble reports.
Well, I think the problem is that providers lock people in 1 and 2 year contracts so if people get bad service they are stuck. That is why I only have month to month service, if people don't like there T1 then they can quit. I think we need more providers to do that, I know of a lot users that are on providers that want to switch, but have 8 months left on their contract.
Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite 5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest ---------------------------------------------------------------------------
Nathan writes:
Well, I think the problem is that providers lock people in 1 and 2 year contracts so if people get bad service they are stuck. That is why I only have month to month service, if people don't like there T1 then they can quit. I think we need more providers to do that, I know of a lot users that are on providers that want to switch, but have 8 months left on their contract.
I think you're confusing "quality" with "locked into contract." If you have quality, being locked in with fixed rates and service is a good thing. If you don't have quality it's a bad thing. Try real hard not to blame the provider, but the idiot who signed a contract that doesn't have performance guarantees. We [ACES Research] offer our clients contracts, and if they want to pay for performance guarantees, those too. None of them whine, so I assume they're happy and not upset. Some of them have locked rates till '98 that are just too good to mention.
Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World!
Ehud -- Ehud Gavron (EG76) gavron@Hearts.ACES.COM
participants (11)
-
Dave Mills -
Ehud Gavron -
Gordon Cook -
hwb＠upeksa.sdsc.edu -
John Jun Wu -
jon＠branch.com -
Michael Dillon -
Mike -
Nathan Stratton -
Rob Skrobola -
Tim Bass