CCing the alias as message went out of bounds in size. On Thu, Feb 7, 2013 at 9:28 PM, Balaji venkat Venkataswami < [email protected]> wrote:
> Dear Eric, > > We seem to be discussing at cross purposes. > > The root of our position is not what you have stated. We do not say if you > dont use a link you consume less power. We are not for avoiding links but > we are for avoiding nodes/routers/switches which have a large power > footprint. The essence of using link metrics is to avoid nodes with larger > power footprints. That is how SPF or CSPF works. When it computes a low > power path it avoids nodes and also links that lead to that node which has > a larger power footprint. So the premise that we are just avoiding links is > WRONG. > > For #1 question, we dont even talk about the power the link draws. We talk > about the power footprint of the node that it is connected to. > > For #2 question, again we state that your premise of avoiding links in > your example is a non-starter for us. > > thanks and regards, > balaji venkat > > > On Thu, Feb 7, 2013 at 9:18 PM, Eric Osborne (eosborne) < > [email protected]> wrote: > >> The root of your position seems to be that if you don't use a link, you >> consume less power, and thus avoiding certain links means consuming less >> power means saving money. >> >> I see two problems with this: >> >> 1) I'm not sure that's actually true. SONET, for one, sends empty frames >> when there's no data, so the laser is always firing. >> >> 2) even if it was true, or if you had some analogue to 802.3az that you >> could use in backbones, I don't believe you'd save anywhere near enough >> money to make it worth it. >> >> For #1, please explain the circumstances under which leaving a link idle >> saves power, and explain how much power and what drives that power draw. >> >> For #2, please see my example in my original email. >> >> >> >> >> eric >> >> >> > -----Original Message----- >> > From: Shankar Raman [mailto:[email protected]] >> > Sent: Thursday, February 07, 2013 4:48 AM >> > To: Balaji Venkat; Eric Osborne (eosborne) >> > Cc: [email protected]; [email protected] >> > Subject: Re: Power aware networks : Comments requested from routing >> > community >> > >> > Dear Eric, >> > Looks like we have not been clear in conveying our idea. >> > >> > We abstract the power consumed by every router/switch, use it as the >> power >> > rating. We then assign it to the link as a metric characteristic to get >> algorithms >> > like SPF and CSPF to choose the lowest power paths or any metric based >> routing >> > algorithm. >> > >> > So the link metric is a combination of power consumed including the >> physical >> > link and router components. In effect the link metric is not just the >> physical >> > link’s metric alone (but the cost of the power consumed by the >> router/switch on >> > which it connects to the neighbor). >> > >> > Just like routing protocols consider delay, hops as metric we consider >> this >> > power consumption as yet another metric. >> > >> > Thanks >> > Shankar >> > >> > -----Original Message----- >> > From: Balaji Venkat >> > Sent: Thursday, February 07, 2013 2:36 AM >> > To: Eric Osborne (eosborne) >> > Cc: [email protected] ; Shankar Raman M J ; [email protected] >> > Subject: Re: Power aware networks : Comments requested from routing >> > community >> > >> > Dear Eric, >> > >> > Comments inline... >> > >> > Sent from my iPad >> > >> > On 07-Feb-2013, at 1:50 AM, "Eric Osborne (eosborne)" >> > <[email protected]> wrote: >> > >> > > You aim to optimize core networks through traffic engineering such >> that >> > some links can be shut down when not in use, or switched to lower power. >> > >> > Actually the answer is a No. We do not advocate this approach. The main >> idea >> > in bgp power path draft has a survey section on previous literature >> that we >> > quote for reference and state that such approaches cause expensive >> equipment >> > to be kept unused. You have quoted the survey section of the paper in >> the >> > above sentence. >> > >> > > >> > > Have you done studies in core networks with real equipment that show >> the >> > actual amount of money you could save with this approach? Let's walk >> through >> > a quick hypothetical network, you tell me where I got it wrong. >> > > >> > > draft-mjsraman-panet-bgp-power-path says " Power consumption can be >> > > reduced by trading off performance related measures like latency. >> For >> > > example, power savings while switching from 1 Gbps to 100 Mbps is >> > > approximately 4 W and from 100 Mbps to 10 Mbps around 0.1 Watts." >> > >> > Again this is a survey section of the paper surveying the existing >> approaches. >> > >> > > >> > > but provides no documentation to back these costs up nor itemization >> > > of where the power is consumed (amortized across the router? linecard >> > > optics? DWDM gear?) >> > >> > Our main intent is to allow traffic to follow low power paths through >> ASs (either >> > through pce like entities or through modifying the bgp path selection >> > algorithm) which consume the least power and possess the required >> bandwidth. >> > By graphing the topology of a set ASs in the immediate neighbourhood >> using >> > as-path-info strands typically for a provider manning multiple ASs >> within the >> > providers own control this mechanism will let large chunks of traffic >> belonging >> > to a macro FEC traverse such low power ASs through an inter-as te MPLS >> lsp >> > computed by a pce like entity and setup by RSVP-te. >> > >> > The use of the metric, the way the pwr metric is arrived at , the >> computation of >> > the topology of ASs, consequent CSPF and then mapping the traffic to the >> > constructed LSP is the main focus if this paper. For dampening >> fluctuations in >> > the metric which are frequent heuristics algorithms are suggested. >> > >> > By no means are we advocating shutting off links. This scheme >> facilitates even >> > the follow the moon strategy naturally. >> > >> > >> > > >> > > Let's assume that this is true and that it is linear, so that you burn >> > > 4W per Gb. (side note: I suspect this is inaccurate and that scaling >> > > is sublinear, but let's go with it because if it's sub-linear you have >> > > even less of a use case. I also think the real world is far more >> > > complex, as you have all the optical transport gear to worry about. >> > > But let's go with it for now.) >> > > >> > > Running a 100Gb link thus draws 400W. >> > > >> > > Let's say you have a backbone with (300) 100Gb links. Total power >> > > consumption is thus (300*400) == 120 kW >> > > >> > > Running all these links for 24 hours thus draws 120*24 == 2,880 kWh >> > > == 2.8mWh Power costs are maybe $0.10/kWh in the US. Double that to >> > cover the cost of cooling, so $0.20/kWh. >> > > Thus, running the entire backbone costs $0.2 * 2,880 = $576/day. >> > $210,000/year. >> > > Let's say your approach can save one third of the power cost, which >> means >> > about $70,000. >> > > >> > >> > >> > >> > > An operator with 300 100Gb links in a network has hundreds of >> millions of >> > dollars worth of gear and millions or tens of millions in payroll >> alone. If you cut >> > $70k from their opex they probably wouldn't even notice. That's one or >> two >> > salaries, or 0.002% of what Time Warner spent on advertising in 2006 >> (see: >> > http://gaia.adage.com/images/random/FactPack06.pdf). It is a drop in >> the >> > bucket, if that. And your proposal comes with significant work >> attached to it, >> > and significant risk. If your 40 years of experience in the network >> industry don't >> > help you understand the risks you're asking an operator to take then >> you're >> > missing a crucial part of any potential real world solution. I do not >> think your >> > work should be presented at the IETF unless it makes a much stronger >> > argument that its benefits outweigh its costs. >> > >> > Again we are not asking the operator to shut off the equipment. He >> needs to >> > keep it running as long as he has traffic to carry except that the pce >> makes a >> > computation every time a large FEC of traffic needs to be carried >> through his >> > ASs through a path that consumes the least power with the links therein >> having >> > the required bandwidth. The ratio of consumed power to the available >> > bandwidth is the metric that is used. Consumed power relates to a >> weighted >> > average of the power consumed within the AS including the edge and core >> > devices while the available bw relates to the links ingressing into the >> AS at the >> > ASBRs that are the inter connectors between the ASes. >> > >> > >> > So in summary, we do not advocate shutting off links at all. In fact we >> disagree >> > with that approach. The scheme we advocate tries to optimise the power >> > consumption using the metric based approach with some heuristics. >> > >> > Hope this helps. >> > >> > Thanks and regards, >> > Balaji venkat >> > > >> > > These sorts of power optimizations all seem to be "here's how you >> reshape >> > the problem so that you can throw a linear problem solver at it". For >> example, >> > [ http://www.cs.berkeley.edu/~suchara/publications/GreenNetsBundles.pdf]. >> > I've never seen anything which shows how much more work it will be for a >> > network operator or which quantifies the actual savings. >> > > >> > > If you can demonstrate significant savings in real networks at little >> or no cost >> > to operators, you have an idea worth pursuing. If your idea will cause >> more >> > operational angst (e.g. not knowing whether your unused capacity will >> be there >> > when you need it because you shut a third of it off all the time, >> increased risk of >> > equipment failure from constant power-cycling, operational tools and >> training >> > and expertise required to manage, deploy and troubleshoot variable-power >> > links and the centralized NMS required to run them, etc etc etc) then >> it will find >> > little traction. Green-TE and power-aware BGP have been floating >> around for a >> > while and have seen no real uptake in the WGs as far as I can see. Is >> that not to >> > be taken as an indication that there may be no real-world interest in >> them? If >> > not, what would it take to convince you? >> > > >> > > >> > > >> > > eric >> >
_______________________________________________ rtgwg mailing list [email protected] https://www.ietf.org/mailman/listinfo/rtgwg
