dslreports logo
site
 
    All Forums Hot Topics Gallery
spc

spacer




how-to block ads


Search Topic:
uniqs
26
share rss forum feed
cramer
Premium
join:2007-04-10
Raleigh, NC
kudos:8

1 recommendation

Re: Physics fail

The key difference being power is produced to match load. If no one is using it, they don't produce it. If they make it and no one uses it, it's generally gone. It takes real work to produce power.

Bits on the other hand are far more transitive. An OC3 is 155Mbps. Either idle pattern or user traffic, it's always moving 155mil bits per second. You cannot store them, or use them later. And by contrast, it takes almost no effort to "create" bits.
dynodb
Premium,VIP
join:2004-04-21
Minneapolis, MN

1 recommendation

Re: Physics fail

said by cramer:

The key difference being power is produced to match load. If no one is using it, they don't produce it. If they make it and no one uses it, it's generally gone. It takes real work to produce power.

Bits on the other hand are far more transitive. An OC3 is 155Mbps. Either idle pattern or user traffic, it's always moving 155mil bits per second. You cannot store them, or use them later. And by contrast, it takes almost no effort to "create" bits.

That's true, but as consumption increases it requires more and more of OC3's (or other such trunks), routers, etc to deliver the bandwidth even if the actual number of customers served remains static. While more indirect, the costs of increased bandwith consumption per user are quite real.

To continue the analogy, many power companies provide incentives to customers to reduce power consumption, especially during peak hours.

Why would a company want you to buy less of their product? The reason is that the cost to build a new power plant or expand the current one is so high, they don't want power consumption to exceed their current capacity to deliver it.
cramer
Premium
join:2007-04-10
Raleigh, NC
kudos:8

Re: Physics fail

While the costs are real, they are also rather trivially low. Sure, it might be a rather large check to sign, but as a function of revenue and functional lifetime, it's peanuts.

Power production, however, is a VERY expensive process. The power companies ask people to reduce PEAK consumption to reduce their need to build facilities to produce power for peak usage -- facilities that would go mostly unused. They don't want you to use less power as much as they want everyone to use a relatively constant level.
dynodb
Premium,VIP
join:2004-04-21
Minneapolis, MN

Re: Physics fail

said by cramer:

While the costs are real, they are also rather trivially low. Sure, it might be a rather large check to sign, but as a function of revenue and functional lifetime, it's peanuts.

Power production, however, is a VERY expensive process. The power companies ask people to reduce PEAK consumption to reduce their need to build facilities to produce power for peak usage -- facilities that would go mostly unused. They don't want you to use less power as much as they want everyone to use a relatively constant level.

Except it's far from trivially low, largely because the functional lifetime is so low due to increased bandwidth demand.

With regards to the peak power analogy, telcos face the same problem. Bandwidth demands are typically low during the day, but increase dramatically evening / night. The same trunks that might only be running at 60% capacity at 4pm may suddenly hit 98%-100% at 9pm. However, they can't really provide incentive for people to shift more of their bandwidth usage to non-peak hours when they're not home to use it.

Instituting a peak / non-peak hours cap would be far more effective, but I'm not sure that it would fly.
cramer
Premium
join:2007-04-10
Raleigh, NC
kudos:8

Re: Physics fail

The hardware lasts a lot longer than you think it does. The existing D2 tech found in many networks has been there for 5+ years. And has seen software upgrades several times. (possibly even soft upgraded from DOCSIS 1.0/1.1) DOCSIS 3.0 is requiring a number of hardware components be upgraded (replaced), but it's not a forklift project -- in many cases the RF line cards are all they have to replace (at least initially.)

And the costs really are peanuts... look at the cost of the gear compared to their monthly profits -- which they've been collecting for YEARS. It's suddenly our fault they haven't planned (and banked) for this day?

ArrayList
netbus developer
Premium
join:2005-03-19
Evanston, IL
Reviews:
·Comcast

Re: Physics fail

said by cramer:

And the costs really are peanuts... look at the cost of the gear compared to their monthly profits -- which they've been collecting for YEARS. It's suddenly our fault they haven't planned (and banked) for this day?

Complete lack of network planning or piss poor network planning on the ISPs part. Upgrading and maintaining your network are the key tenets to running a successful ISP. Without either of those you are done for.
BlueC

join:2009-11-26
Minneapolis, MN
kudos:1
said by dynodb:

That's true, but as consumption increases it requires more and more of OC3's (or other such trunks), routers, etc to deliver the bandwidth even if the actual number of customers served remains static. While more indirect, the costs of increased bandwith consumption per user are quite real.

To continue the analogy, many power companies provide incentives to customers to reduce power consumption, especially during peak hours.

Why would a company want you to buy less of their product? The reason is that the cost to build a new power plant or expand the current one is so high, they don't want power consumption to exceed their current capacity to deliver it.

I think the important thing to mention is the differences between technologies (ADSL/VDSL vs. DOCSIS vs. Wireless vs. Ethernet).

Upgrading capacity on any of these choices is going to vary, but you have to understand that there is a difference between capacity to the core, capacity to the node, and capacity to the end user.

Any of these links can be oversubscribed. How do we know certain ISPs aren't oversubscribing things beyond a reasonable level? Since residential service pretty much comes with no guarantee, they can do whatever they want regarding dedicating a specific capacity to a node (or core network).

Maybe we should ask, "why would they?". Reduced costs.

So, when it comes to upgrading capacity, it can be very easy to do (cost-wise) depending on the technology being used.

Comparing power to bandwidth is foolish. Two completely different services, with VERY different variables.
dynodb
Premium,VIP
join:2004-04-21
Minneapolis, MN

Re: Physics fail

said by BlueC:

I think the important thing to mention is the differences between technologies (ADSL/VDSL vs. DOCSIS vs. Wireless vs. Ethernet).

Upgrading capacity on any of these choices is going to vary, but you have to understand that there is a difference between capacity to the core, capacity to the node, and capacity to the end user.

Any of these links can be oversubscribed. How do we know certain ISPs aren't oversubscribing things beyond a reasonable level? Since residential service pretty much comes with no guarantee, they can do whatever they want regarding dedicating a specific capacity to a node (or core network).

Maybe we should ask, "why would they?". Reduced costs.

So, when it comes to upgrading capacity, it can be very easy to do (cost-wise) depending on the technology being used.

Comparing power to bandwidth is foolish. Two completely different services, with VERY different variables.

Virtually every aspect of the Internet is oversubscribed. Every ISP, telco, cable and wireless provider oversubscribes.

If the amount of bandwidth per user remained static month to month and quarter to quarter, getting the right oversubscription ratio wouldn't be difficult to determine and meet. As it stands now though, that's not the case- it's increased quite dramatically in a relatively short period of time.

Expanding capacity isn't "very easy" with regards to cost or time, especially when you're looking at tens of thousands of trunks, nodes, etc for a major provider. We're talking not millions, but hundreds of millions of dollars for upgrades that may get overwhelmed with increased traffic less than a year later.
BlueC

join:2009-11-26
Minneapolis, MN
kudos:1

Re: Physics fail

said by dynodb:

Virtually every aspect of the Internet is oversubscribed. Every ISP, telco, cable and wireless provider oversubscribes.

If the amount of bandwidth per user remained static month to month and quarter to quarter, getting the right oversubscription ratio wouldn't be difficult to determine and meet. As it stands now though, that's not the case- it's increased quite dramatically in a relatively short period of time.

Expanding capacity isn't "very easy" with regards to cost or time, especially when you're looking at tens of thousands of trunks, nodes, etc for a major provider. We're talking not millions, but hundreds of millions of dollars for upgrades that may get overwhelmed with increased traffic less than a year later.

Absolutely, but there are certain parts that should not be, or it should be done with very little oversubscription. i.e. the connection servicing the node. You already have substantial oversubscription at the node in correlation with capacity delivered to the end user, if you oversubscribe the capacity to the node, it becomes exponential.

That's where things get messy. But hey, it saves the ISP money!

I said expanding capacity CAN be easy, depending on the scenario. If I wanted to introduce an additional GigE to my core, it would be very easy to accomplish, and not all that expensive. The key thing is I planned for future upgrades. Most ISPs do, so if they're excuse is they can't add additional capacity, then they didn't plan accordingly or they're wanting to save money.

Comcast seemed to have no trouble going from DOCSIS 2.0 -> 3.0. That was a big upgrade, yet they were quick to start deploying it in their markets (compared to other ISPs, I am in no way saying what they did was quick in general).

Of course demand has been rising substantially over the years, but so has the decrease in cost of bandwidth. There have been numerous advancements in technology which have allowed for greater increases in capacity without having it cost an arm and a leg.

DOCSIS 3.0 is a prime example. It's allowed ISPs to make capacity upgrades quite easily.
dynodb
Premium,VIP
join:2004-04-21
Minneapolis, MN

Re: Physics fail

said by BlueC:

I said expanding capacity CAN be easy, depending on the scenario. If I wanted to introduce an additional GigE to my core, it would be very easy to accomplish, and not all that expensive. The key thing is I planned for future upgrades. Most ISPs do, so if they're excuse is they can't add additional capacity, then they didn't plan accordingly or they're wanting to save money.

Comcast seemed to have no trouble going from DOCSIS 2.0 -> 3.0. That was a big upgrade, yet they were quick to start deploying it in their markets (compared to other ISPs, I am in no way saying what they did was quick in general).

Adding a GigE to one core using existing infrastructure? OK, not a big deal. Adding a GigE to 10,000 cores? Different story.

And yes, the techology used makes a difference, but it's also expensive and time consuming to upgrade technologies- for instance going from DSLAMs fed by ATM DS3s or OC3s to those fed by GigE fiber. If, for example, the DSLAM isn't capable of using a GigE trunk, now not only do you have to install ethernet switches and fiber, but install a new DSLAM to use it.
BlueC

join:2009-11-26
Minneapolis, MN
kudos:1

Re: Physics fail

said by dynodb:

Adding a GigE to one core using existing infrastructure? OK, not a big deal. Adding a GigE to 10,000 cores? Different story.

And yes, the techology used makes a difference, but it's also expensive and time consuming to upgrade technologies- for instance going from DSLAMs fed by ATM DS3s or OC3s to those fed by GigE fiber. If, for example, the DSLAM isn't capable of using a GigE trunk, now not only do you have to install ethernet switches and fiber, but install a new DSLAM to use it.

It's all relative.

If you have 10,000 cores, you probably have a ton of customers to make up for the cost. You also have more leverage with the carriers you're doing business with to get those said GigE connections. So in the end, the cost per user is substantially less than dealing with a network that has a lower volume.

rchandra
Stargate Universe fan
Premium
join:2000-11-09
14225-2105
I would say that's true at the line level, but anything beyond requires a delta of power to process. What I'm talking about is the power for example to run a router's CPU to take that data from the one link, make a decision about its proper destination, and optionally (if firewalling policy permits) copy it to another interface, to a DSLAM, to a CMTS, etc. I don't have any numbers on what that delta is for any given router, but I can virtually guarantee you it's nonzero.

I was happy to contribute CPU cycles to SETI until I calculated how much more juice my computer was taking.
--
English is a difficult enough language to interpret correctly when its rules are followed, let alone when a writer chooses not to follow those rules.


Jeopardy! replies and randomcaps REALLY suck!