RE: Is anyone actually USING IP QoS?
Hmm. It was interesting to read the various replies both on the list and personal. A number of people pointed out, rightly, that I'm thinking of the USA. That's true. They further gave examples of the exhorbitant costs that people elsewhere in the world pay for connections. While I certainly understand the desire to use some form of QoS to squeeze every bit of capacity from expensive skinny pipes, I believe that it is wrong to try to use a technological solution (usually poorly understood and implemented) to solve what is primarily a political problem. There is absolutely no technological reason why international rates are so much higher than domestic US rates. Regarding the idea of "free bandwidth," that's not what I said. I said that bandwidth is "essentially free." Of course there will always be a cost for bandwidth. But consider for a moment what's happened to disk storage. In 1990 I purchased my first PC. I paid $550 for an 80 MB hard drive -- that's $6.875 per megabyte. Today you can purchase a 25 GB hard drive for $450 -- that's $0.018 per megabyte. That's a 31,250% increase in capacity accompanied by a 99.73% reduction in price per megabyte. So you see, on a per megabyte basis, storage is "essentially free." The same thing has happened to CPU and memory. It will happen to bandwidth, too, and in many cases already has. For a time I worked in information technology architecture. One of the tennents of that field is that it's always cheaper to trade capacity for staff. You can overbuild now, while planning for growth, and save money over the alternative of continually tweaking and making minor improvements and upgrades which requires expensive time and personnel. QoS is an acceptable idea which in certain specific situations might be suitable for solving an existing problem. But since QoS is expensive to manage it simply is not viable in the long term. We in the networking and telecommunications industries need to redirect our energy away from bandages and instead toward making abundant bandwidth readily available to everyone. _________________________________________________________ Steve Riley Microsoft Telecommunications Practice in Denver, Colorado email: mailto:steriley@microsoft.com call: +1 303 521-4129 (cellular) page: +1 888 440-6249 or mailto:4406249@skytel.com Applying computer technology is simply finding the right wrench to pound in the correct screw. -----Original Message----- From: Vadim Antonov [mailto:avg@kotovnik.com] Sent: Monday, May 17, 1999 4:29 PM To: nanog@merit.edu; Steve Riley (MCS) Subject: RE: Is anyone actually USING IP QoS? Steve Riley (MCS) <steriley@microsoft.com> wrote:
Nice to see that I'm not the only one believing in the foolishness of QoS hype.
Er... me, agreeing with someone from Microsoft? Yeech! (just kidding :)
Allow me to point you to an interesting paper called "Rise of the Stupid Network."
Unfortunately that paper oversimplifies the whole congestion control issue by completely ignoring the fact that data traffic has a heavy-tailed distribution. Which pretty much means that no matter how much capacity is there, as long as there's oversubscription there will be at least transient traffic jams. Which means that the issue of what to do with different types of traffic when there's a congestion cannot be just pooh-poohed. The big problem with G.711 and its progeny is the lack of effective cooperative congestion control which would guarantee network stability (like TCP does). Fortunately, the bulk of network traffic is "canned" content, which can (and should) be transmitted with TCP. --vadim
steriley@microsoft.com said:
There is absolutely no technological reason why international rates are so much higher than domestic US rates.
?? Last time I looked the cost of transatlantic / transpacific cable laying (per strand) was rather more than the cost of land-based cables. There are lots of reasons for this which are primarilly technology based (like what equipment works under water). There are actually 2 differences: Firstly political ones to do with circuits crossing country borders, and who is paying for the circuit, and then the ones to do with geography and resultant tecnological factors. Intra-European bandwidth now appears to be as commoditized (if not more) than intra-US bandwidth. Certainly there are *more* suppliers of European IRU and dark fiber with live or soon-to-be-live network (provided you only want to go to London, Amsterdam, Paris & Frankfurt) than (say) you can buy NY, WDC, SF, LA. However, removing political obstacles and introducing competition does not also remove the atlantic. -- Alex Bligh GX Networks (formerly Xara Networks)
There is absolutely no technological reason why international rates are so much higher than domestic US rates.
so, as a pragmatist, should i spend my time telling the world they're wrong or getting my job of moving packets done?
Regarding the idea of "free bandwidth," that's not what I said. I said that bandwidth is "essentially free." Of course there will always be a cost for bandwidth.
what percentage of a stateside isp's p/l do you think circuit cost is? what percentage of a european isp's p/l do you think circuit cost is? what percentage of an african isp's p/l do you think circuit cost is? hint: the numbers do not support your argument. some day they may. in the meantime, do you mind if no one delivers packets to you until that day comes? randy
This conversation has been very useful, but gotten a little bit away from what I was originally asking. Though it has still been interesting. Can I redirect the discussion back towards whether IP QoS/CoS is capable of replacing ATM in a multi-service network, limited to the boundaries of your own network? If I were to put together a native-IP network (probably POS) which was running high levels of traffic (at least DS-3 to start and probably OC-3), and put services such as dial-up, hosting/co-locate, DSL, and high-speed dedicated, and mix in VoIP, is it feasible to use one IP pipe and logically segment it with IP QoS/CoS features? So far, it seems that the way to do this is to physically or logically segment the network using separate circuits or ATM. But it would make a lot of sense (in theory at least) to use IP and bunch everything together. Pete.
If I were to put together a native-IP network (probably POS) which was running high levels of traffic (at least DS-3 to start and probably OC-3), and put services such as dial-up, hosting/co-locate, DSL, and high-speed dedicated, and mix in VoIP, is it feasible to use one IP pipe and logically segment it with IP QoS/CoS features?
You just described our network, and what several of our customers are doing. In other words, Yes. Now, it's not for blue haired ladies or non-innovative IT directors [translate to low risk tolerance]. But it absolutely is done today.
So far, it seems that the way to do this is to physically or logically segment the network using separate circuits or ATM. But it would make a lot of sense (in theory at least) to use IP and bunch everything together.
Bah. Keep listening to your ATM vendors... -a
"Steve Riley (MCS)" wrote:
Regarding the idea of "free bandwidth," that's not what I said. I said that bandwidth is "essentially free." Of course there will always be a cost for bandwidth. But consider for a moment what's happened to disk storage. In 1990 I purchased my first PC. I paid $550 for an 80 MB hard drive -- that's $6.875 per megabyte. Today you can purchase a 25 GB hard drive for $450 -- that's $0.018 per megabyte. That's a 31,250% increase in capacity accompanied by a 99.73% reduction in price per megabyte. So you see, on a per megabyte basis, storage is "essentially free." The same thing has happened to CPU and memory. It will happen to bandwidth, too, and in many cases already has.
I'll concede this point. However, it's worth pointing out that megabytes of disk space can actually be 'had', while bandwidth doesn't really exist. Most of us, at least, have to pay for bandwidth by the month, while we only pay for the same piece of storage space once (and then we actually posess it). This is important because it impacts how far this cost model can be extended, and for how long. -- Nick Bastin - RBB Systems, Inc. The idea that Bill Gates has appeared like a knight in shining armour to lead all customers out of a mire of technological chaos neatly ignores the fact that it was he who, by peddling second-rate technology, led them into it in the first place. - Douglas Adams
In message <374268F7.C82ED3DA@rbbsystems.com>, Nick Bastin writes:
"Steve Riley (MCS)" wrote:
Regarding the idea of "free bandwidth," that's not what I said. I said that bandwidth is "essentially free." Of course there will always be a cost for bandwidth. But consider for a moment what's happened to disk storage. In 1990 I purchased my first PC. I paid $550 for an 80 MB hard drive -- that's $6.875 per megabyte. Today you can purchase a 25 GB hard drive for $450 -- that's $0.018 per megabyte. That's a 31,250% increase in capacity accompanied by a 99.73% reduction in price per megabyte. So you see, on a per megabyte basis, storage is "essentially free." The same thing has happened to CPU and memory. It will happen to bandwidth, too, and in many cases already has.
I'll concede this point. However, it's worth pointing out that megabytes of disk space can actually be 'had', while bandwidth doesn't really exist. Most of us, at least, have to pay for bandwidth by the month, while we only pay for the same piece of storage space once (and then we actually posess it). This is important because it impacts how far this cost model can be extended, and for how long.
Its also worth pointing out, that a lot of us are paying for bandwidth we aren't using, but this doesn't impact the validity of the original statement. Bandwitdh costs and bandwidth availaablity have significantly dropped in costs over the last 4 years. I can remeber $8000/month T-1s (for Internet), and I'm now paying less for long haul DS3s that AT&T was quoting me for T-1s. If we measure the projected 10 year price drop on bandwidth in a linear fashion the $5.33/kbit/month goes to $.40/kbit/month. The $.55/ds0/mile to less than $.03/ds0/mile. Now, one also has to consider that the price per megabit for storage on 40meg hard drives, has not dropped much in the last 4 years or so, but in 2 gig drives it has. I used to work at a company that is now in the top 5 computer makers, and there is a minimum price at which it is interesting to sell a product that decreases slower than the price per megahertz/bit/whatever. All the counter examples to the countrary, if you look at the projected sales of just a few of the players in the lang haul fiber biz, you can quickly compare that to computer maker volumes, and get a REAL good guess as to what the price per ds0/mile will be in 5 years. And the cost of bandwidth (IP transport) compared to local loop costs (also considering Europe/AISA has having really long local loops, not saying its right, fair, or will stay that way), the cost really starts to become insignficant for burstable services, which is what end users services are, (video on demand, VoIP, etc, etc, etc.) The generalization that bandwidth is estentially free, is true, as far as generalizations go, or will be withing a few years, when you consider the other real costs of doing business. And the volumes of data for point to point, or point to multipoint, voice and video conferencing are so completely trivally compared to completely replicatable and cachable content, it is just silly. QoS outside of a private network is not ever going to be an economic reality, unless an RBOC buys up all the telephone companies or something equally stupid. (Its happened before.) I'm not even convinced that from an engineering standpoint it isn't going to be cheaper for people with "short" local loops, to just always buy more capacity, than, ever drop any significant number of packets. --- jerry@fc.net Insync Internet, Inc. | Freeside Communications, Inc. 5555 San Felipe, Suite 700 | PO BOX 80315 Austin, Tx 78708 713-407-7000 | 512-458-9810 http://www.insync.net | http://www.fc.net
Jeremy Porter wrote:
QoS outside of a private network is not ever going to be an economic reality, unless an RBOC buys up all the telephone companies or something equally stupid. (Its happened before.)
Um, there are a number of "public" networks delivering QoS today--they just happen to be using layer 2 technology. Many of your telephone calls, in your part of the country and others, are transported by "public" data networks, and have been for years. You might not know, nor should you care. It just works. Many major sporting events are also transported over public data networks. The underlying technology now exists to make this QoS capability available with "public" Layer 3 networks. It will have significant bugs at first--just as Layer 2 networks initially did. But it will eventually become stable enough that businesses depend upon it. The economic justification is in charging your customers more for a higher level of QoS. It's been done before--just not for the Internet, yet.
I'm not even convinced that from an engineering standpoint it isn't going to be cheaper for people with "short" local loops, to just always buy more capacity, than, ever drop any significant number of packets.
This is probably true. Just as bandwidth within an enterprise is cheap, short local loop bandwidth will also be cheap for those with easy access to facilities. It's the long haul where the real economic gains are.
--- jerry@fc.net Insync Internet, Inc. | Freeside Communications, Inc. 5555 San Felipe, Suite 700 | PO BOX 80315 Austin, Tx 78708 713-407-7000 | 512-458-9810 http://www.insync.net | http://www.fc.net
---------------------------------------------------------------------- Prabhu Kavi Phone: 978-264-4900 x125 Tenor Networks FAX: 978-264-0671 50 Nagog Park Email: prabhu_kavi@tenornetworks.com Acton, MA 01720 WWW: www.tenornetworks.com ----------------------------------------------------------------------
Hmm. Let me one more word. Very short. And I promise to be quiet anymore. When I order bandwidth, I got BANDWIDTH. Just what I asked. When I install RSVP and QoS software tricks, what will I have? Nothing predictable - it can work, it can not work, it can work for months and then destroy itself. The density of bugs increase every months (true for CISCO, true for MS, I think it's true for other vendors). Result? How can sales people use something mistical? They prefer to get solid and simple way - order bandwidth. On Tue, 18 May 1999, Steve Riley (MCS) wrote:
Date: Tue, 18 May 1999 13:02:28 -0700 From: Steve Riley (MCS) <steriley@microsoft.com> To: nanog@merit.edu Subject: RE: Is anyone actually USING IP QoS?
Hmm. It was interesting to read the various replies both on the list and personal. A number of people pointed out, rightly, that I'm thinking of the USA. That's true. They further gave examples of the exhorbitant costs that people elsewhere in the world pay for connections. While I certainly understand the desire to use some form of QoS to squeeze every bit of capacity from expensive skinny pipes, I believe that it is wrong to try to use a technological solution (usually poorly understood and implemented) to solve what is primarily a political problem. There is absolutely no technological reason why international rates are so much higher than domestic US rates.
Regarding the idea of "free bandwidth," that's not what I said. I said that bandwidth is "essentially free." Of course there will always be a cost for bandwidth. But consider for a moment what's happened to disk storage. In 1990 I purchased my first PC. I paid $550 for an 80 MB hard drive -- that's $6.875 per megabyte. Today you can purchase a 25 GB hard drive for $450 -- that's $0.018 per megabyte. That's a 31,250% increase in capacity accompanied by a 99.73% reduction in price per megabyte. So you see, on a per megabyte basis, storage is "essentially free." The same thing has happened to CPU and memory. It will happen to bandwidth, too, and in many cases already has.
For a time I worked in information technology architecture. One of the tennents of that field is that it's always cheaper to trade capacity for staff. You can overbuild now, while planning for growth, and save money over the alternative of continually tweaking and making minor improvements and upgrades which requires expensive time and personnel. QoS is an acceptable idea which in certain specific situations might be suitable for solving an existing problem. But since QoS is expensive to manage it simply is not viable in the long term. We in the networking and telecommunications industries need to redirect our energy away from bandages and instead toward making abundant bandwidth readily available to everyone.
_________________________________________________________ Steve Riley Microsoft Telecommunications Practice in Denver, Colorado email: mailto:steriley@microsoft.com call: +1 303 521-4129 (cellular) page: +1 888 440-6249 or mailto:4406249@skytel.com Applying computer technology is simply finding the right wrench to pound in the correct screw.
-----Original Message----- From: Vadim Antonov [mailto:avg@kotovnik.com] Sent: Monday, May 17, 1999 4:29 PM To: nanog@merit.edu; Steve Riley (MCS) Subject: RE: Is anyone actually USING IP QoS?
Steve Riley (MCS) <steriley@microsoft.com> wrote:
Nice to see that I'm not the only one believing in the foolishness of QoS hype.
Er... me, agreeing with someone from Microsoft? Yeech! (just kidding :)
Allow me to point you to an interesting paper called "Rise of the Stupid Network."
Unfortunately that paper oversimplifies the whole congestion control issue by completely ignoring the fact that data traffic has a heavy-tailed distribution. Which pretty much means that no matter how much capacity is there, as long as there's oversubscription there will be at least transient traffic jams.
Which means that the issue of what to do with different types of traffic when there's a congestion cannot be just pooh-poohed.
The big problem with G.711 and its progeny is the lack of effective cooperative congestion control which would guarantee network stability (like TCP does). Fortunately, the bulk of network traffic is "canned" content, which can (and should) be transmitted with TCP.
--vadim
Aleksei Roudnev, Network Operations Center, Relcom, Moscow (+7 095) 194-19-95 (Network Operations Center Hot Line),(+7 095) 230-41-41, N 13729 (pager) (+7 095) 196-72-12 (Support), (+7 095) 194-33-28 (Fax)
participants (9)
-
Alan Hannan -
Alex Bligh -
Alex P. Rudnev -
Jeremy Porter -
Nick Bastin -
Pete Kruckenberg -
Prabhu Kavi -
Randy Bush -
Steve Riley (MCS)