& cplSiteName &
Comments
Newest First  |  Oldest First  |  Threaded View        ADD A COMMENT
<<   <   Page 15 / 18   >   >>
sgan201
sgan201
12/5/2012 | 4:08:30 AM
re: QOS Fees Could Change Everything
Seven,

People are allocated 20Kbps. But, it is not guaranteed. How much would it costs to make those 20Kbps GUARANTEED as opposed to only ALLOCATED?? My contention is it does not costs that much. I see that your believe it is hard to do...

You cannot sell something that is allocated as premium service but you can sell something that is guaranteed as premium service. In either case, it uses the same amount of bandwidth.

As a consumer, I find it unbelievable that an ISP will want to sell me all kind of premium services but it cannot even guarantee a minimum level bandwidth to Internet.

It is your belief that ISP and access provider have a choice and it is not worthwhile for ISP/ Access provider to be better bit providers. Perhaps your are right. But, somehow, in my gut feeling, I feel that the genie is out of the bottle. It will be commoditized and whoever is prepared to be a good commodity bit provider will win.

Dreamer
rjs
rjs
12/5/2012 | 4:08:29 AM
re: QOS Fees Could Change Everything
You may actually get it!

Message 141
" ..... The days of commoditization of bit transporting are nearing an end I would say. "


Imagine a US without the commoditization of gas and oil and food. Every one of these vendors wanting to be vertically integrated to get a better value.

Very often one has to take a broader look. Mixing
State and Religion makes sense at many a level, but it is really not a good idea, no matter how many arguments we hear for it.

Common Carriage laws are the ONLY logical way and
bit transport has to be commoditized. There has to be a separation of transport and services.

These things come back to haunt you and there are not easy fixes here.

-RJS




stephencooke
stephencooke
12/5/2012 | 4:08:29 AM
re: QOS Fees Could Change Everything
Mark,

"It was not my impression that IP/Ethernet equipment could only operate at 30% of a 10Gbps. I wonder if this is all equipment or just some? Would be interesting to hear what experience led you to this understanding and what type of equipment (if you don't want to name specific vendors I would understand)."

I can't put names to this but here is a 50kft pass at it:
- your 10/100MB/s network card in your PC has an elastic store that buffers between the main processor card and the LAN interface. The chip that interfaces to the line (lets talk CAT5 cable here) has the capability of running at 100MB/s continually. However, the software that runs that card (ie: the driver) uses the same clock. Can you see the issue? As long as the communications processing can be done in H/W alone, you can maintain the specified bit rate. Once any S/W has to do a memory access to decide anything the H/W has to insert wait states. Therefore the actual data transfer rate is much less than the maximum line rate.

- Telecom networks were all based on the same traffic type and bandwidth (64 kb/s). On call setup a message is sent the entire length of the transmission path to the destination, configuring the switch fabrics along its way for both forward and reverse paths. Obviously this is a 'connection oriented' system. Once the call was setup there was essentially a digital 'wire' between the points involved and all processing was entirely H/W-based so the line rate could be fully utilized.

- The highest clock rate in transmission equipment and routers is the line rate of the interfaces. That is divided by at least 8 (to get byte-oriented words) for processing. The chips in use today combine a bunch of functions on a single piece of silicon. What this means is that the chips that figure out where the byte boundaries are in a serial stream also do some of the routing function (eg: this packet goes to the card in slot x. The division of the overall routing function to pieces of H/W is totally vendor-dependent). As long as the on-chip memory has been setup (either by recent previous packets to the same address, or via a connection set-up function) and the re-directing can be entirely done via H/W alone, the line rate can be maintained. Once S/W gets involved (eg: a decision has to be made or a new address has to be looked up) or there is congestion on a link, the actual data throughput falls below the line rate.

- The firewall/router that you have in your home is likely a 10/100 box that has a processor that runs at <5MB/s and my be less powerful than a 486. Consider what happens when it is decrypting VPN data or doing deep packet inspection.

Bottom line is that comparing telecom bandwidths with actual IP data throughput numbers is not a fair fight. Unless vendors put beefier (ie: faster than line rate) processors on their boxes transmission efficiencies will not significantly improve from what they are now, as far as I can see.

Of course I am an old optical systems guy who knows relatively little about data routing or any of the names that you and PO talk about. I hope this helps.

Steve.
Mark Seery
Mark Seery
12/5/2012 | 4:08:28 AM
re: QOS Fees Could Change Everything
Steve,

Thanks for your response, that clears things up. I am not an expert on end station performance, but your points seem plausible. In terms of router performance, I am aware of some designs where all the traffic forwarding is done in hardware (and ternary CAMs have enough capacity for all network prefixes). I understand your point about the variability of implementation.

In terms of me and PO, all you really have to know is the the following:

There is a natural tension in many designs and architectures (technology, political, etc.) between maximum system (in this case network)evolvability (decentralizing traffic mananagement and services to the end points) and optimizing (economically, business model, etc.) a network for any given set of assumed requirements and services - this has been clearly stated even by Dr. Reed (co-author of end to end argument) who would appear to be an engineer as opposed to a sectarian. The former also requires a trust model (everyone uses TCP, new congestion management proposals for UDP, or UDP+an application level capability that is "fair" to TCP traffic) while the later often assumes that the trust model is not scaleable /trustable and therefore it is up to the operator to embed certain protections and mechanisms within the network (for example QoS) - both as a function of performance and of business model / value proposition. These are two different and respected ways of looking at communications networks, and both philosphical threads can be seen inside and outside the Internet community (see for example comments from Mr. Postel in IEN2 about how TCP should just be a host end to end congestion management, that led to the separation of TCP and IP). With respect to the latter model, it should also be noted that one end to end technology like MPLS is not required (though that may provide operational efficiencies) because per hop / per service / per flow / per aggregate behaviours can be implemented with any number of technologies. I expect that both models will exist together for a long time to come, even if in some cases one is an overlay on the other; I also believe that good architectural thinking can lead to a degree of network evolvability even in the latter model.

all the other "names" are just two people pretending they know what they are talking about and can talk in more precision than is actually likely IMO given the non uniform nature of network topologies, capacities, implementations, and measurements (for example consider the problem of distributed scheduling in an even well characterized cross bar as a starting point); my advise is to not pay too much attention - nothing wrong with being an "old optical systems guy" - there are multiple respected models of "transport" each with their own strengths and weaknesses; it should be a market place of competing ideas and value propositions; not a tyranny of thought.
stephencooke
stephencooke
12/5/2012 | 4:08:28 AM
re: QOS Fees Could Change Everything
RJS,

Who said I wished for this...? I was just extrapolating from the announcement in the source article and some points that Seven made.

"There has to be a separation of transport and services."

Consider the guy who is the VP Sales of transport in an RBOC. His CEO is telling him that he has to start generating more revenue and his personal review is based on hitting new and improved targets. His take would likely be that transport of bits is a service.

"These things come back to haunt you and there are not easy fixes here."

I am willing to be proven wrong here. Any other prognostications? I agree that there are no easy fixes once this is started.

Steve.
paolo.franzoi
paolo.franzoi
12/5/2012 | 4:08:28 AM
re: QOS Fees Could Change Everything

stephen,

The start of the QoS stuff will be video networks (aka the IPTV networks). The DSCPs will be set by the Set Top Boxes.

Eventually, there will have to be a hardware key to set the QoS at a consumer site. This is to allow these to be trusted. In the direction towards the consumer, the VPN connection itself will imply the QoS.

I would expect that these data hotels will exist in several points on the network. It may not be as much traffic as you think, especially at first. Of course, as the traffic grows the revenue is there to justify any network expansion.

And yes, you have found the flaw with QoS over the Internet. Basically, until a common scheme is adopted on every piece of equipment on the Internet it will not work.

Finally, the independent ISPs add no value to this equation. The Cable companies do not support them. The RBOCs are bigger in the broadband arena than any ISP. So, it would be simpler for a company to set up a relationship with the 4 RBOCs and the 3 largest cable companies than it would be with any other provider set.

seven
paolo.franzoi
paolo.franzoi
12/5/2012 | 4:08:27 AM
re: QOS Fees Could Change Everything

spelurker,

That method works until the moment video is deployed. At that moment, the DSLAM becomes a congested element in the worst case. Thus, all those fancy deep inspection boxes occur after the congestion.

seven
spelurker
spelurker
12/5/2012 | 4:08:27 AM
re: QOS Fees Could Change Everything
> Given that the carriers are selecting the home gateways that
> they deploy (I am assuming here that these gateways will
> support DSCP), this will also imply that the applications
> will also have to support DSCP to make use of this capability

Actually, that is not stricly true. Companies like Sandvine and Ellacoya provide service providers with boxes that tag packets with QoS markings based on 'deep packet inspection'. (this occurs on the network ingress, just after DSLAM/CMTS processing) No service provider in their right mind is going to trust the PC in someone's home to properly identify data streams.

> If the entire network supports [Differentiated Services]
> won't the general Internet traffic be assigned a default
> (ie: low) priority level?

Yup. That's the way the network works today. The RBOCs are (supposedly) now offering a way for 3rd parties to make use of the prioritization, which should increase the percentage of traffic which is prioritized. Therefore the BW available for everyone else should shrink proportionately. It actually shouldn't be too much of a problem, since the premium traffic will be there, whether it is prioritized or not.
stephencooke
stephencooke
12/5/2012 | 4:08:27 AM
re: QOS Fees Could Change Everything
Spelurker,

"No service provider in their right mind is going to trust the PC in someone's home to properly identify data streams."

If the carrier has implemented an SLA with the consumer regarding the amount of traffic they are allowed at a given priority level (and presumably the price they will pay for exceeding this level), why would they care? Presumably the DSLAM, or more likely the home gateway, would be doing the policing of the SLA...?

I guess this brings up another point: ownership of the home gateway. If the consumer owns the home gateway then the only way forward is as you have described. If the carrier owns the gateway it can do the policing and prioritization prior to the traffic hitting the network. IMO it is better for the carrier to own the CPE for the above and other reasons.

Steve.
stephencooke
stephencooke
12/5/2012 | 4:08:26 AM
re: QOS Fees Could Change Everything
Spelurker,

"Packet pushing boxes may use, say 10Gb/s on their I/O, but almost always have a significantly faster switching fabric. (A POS interface is serial, but while a router's internals might run at only 500MHz clock rate, they're passing data 32bits wide, giving them 16Gb/s of processing capability)"

Thanks. My question is can the routers, any of them, run at 100% utilization on all of their ports, all the time?

"In a core box, forwarding is 100% hardware, and usually deterministic, because all possible forwarding decisions have been set up based on IP destination + policy constraints."

Assuming the above, why is it not totally deterministic? Regarding your comment on all possible forwarding decisions...how can this be true when it is possible to physically move IP addresses, not just to different stationary equipment but to mobile devices (ie: an address could theoretically move from one port to another, to another, to bypassing that router all together)?

"So with today's IP boxes and LAN/WAN topologies, the effective line utilization gets MUCH higher than we were all taught was possible during the 10base-T days."

Cool. I was under the impression that the 10base-T max utilizations were on the order of 10-15%. What do you feel the current ones are?

Thanks,

Steve.
<<   <   Page 15 / 18   >   >>


Featured Video
Upcoming Live Events
November 5, 2019, London, England
November 7, 2019, London, UK
November 14, 2019, Maritim Hotel, Berlin
December 3-5, 2019, Vienna, Austria
December 3, 2019, New York, New York
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events