Re: [address-policy-wg] 2005-01 - Last Call for Comments (HD-ratio Proposal)
Hi Geoff,
I was also surprised by this number [46%] when I first saw it in the output.
Your number is higher, but the analysis I did also showed HD ratio could have a significant impact on the address space consumption. (http://www.ripe.net/ripe/policies/proposals/comments/impact_of_hd.html posted on this list some weeks ago) Looking at all invidual allocations done by RIPE NCC between 2003 and 2006, we modelled the observed growth to a policy which used HDR 0.96 instead of 80% utilisation as the criterium for an LIR to be eligible to receive an additional allocation. Starting 1/1/2003 and stepping through time the simulation thus determined the address space held by each LIR on a day by day basis. By 1/1/2006 this resulted in some 60 million (about 30% of the total) more addresses allocated compared to what we actually had handed out under the 80% rule. Reading your report, I believe one of the reasons our numbers differ is that you are simulating 10,000 allocations; my analysis only looked at the 5,121 allocations done by RIPE NCC in 2003-2006. Since the effects of HD ratio are progressive, the more allocations you simulate, the higher the relative increase in address space consumption becomes.
This experiment has been repeated 1,000 times in order to determine a stable average value for the relative increase in address consumption corresponding to a change in the address allocation policies from uniform 80% to an HD Ratio of 0.96, assuming constant demand for addresses.
To get a feeling of how stable your average is, could you indicate what the variation, the standard deviation in these 1,000 repeats is? i.e. did all 1000 give you a number close to 46% or were they spread out a lot?
A related consideration is that of the adoption of such a policy proposal by all 5 RIRs.
From http://www.arin.net/policy/proposals/proposal_archive.html I understand ARIN already abandoned two proposals to use HD ratio for IPv4 allocations (nrs. 2004-2 and 2003-10).
Regards, -- Rene =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Rene Wilhelm RIPE Network Coordination Centre Email: wilhelm@ripe.net Amsterdam, the Netherlands Phone: +31 20 535 4417 Fax: +31 20 535 4445 =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Have either of you run the simulations with other HDR values? Would .97 make a significant difference?
-----Original Message----- From: address-policy-wg-admin@ripe.net [mailto:address-policy-wg- admin@ripe.net] On Behalf Of Rene Wilhelm Sent: Wednesday, February 22, 2006 5:44 PM To: Geoff Huston Cc: Randy Bush; address-policy-wg@ripe.net Subject: Re: [address-policy-wg] 2005-01 - Last Call for Comments (HD- ratio Proposal)
Hi Geoff,
I was also surprised by this number [46%] when I first saw it in the output.
Your number is higher, but the analysis I did also showed HD ratio could have a significant impact on the address space consumption. (http://www.ripe.net/ripe/policies/proposals/comments/impact_of_hd.html posted on this list some weeks ago)
Looking at all invidual allocations done by RIPE NCC between 2003 and 2006, we modelled the observed growth to a policy which used HDR 0.96 instead of 80% utilisation as the criterium for an LIR to be eligible to receive an additional allocation. Starting 1/1/2003 and stepping through time the simulation thus determined the address space held by each LIR on a day by day basis. By 1/1/2006 this resulted in some 60 million (about 30% of the total) more addresses allocated compared to what we actually had handed out under the 80% rule.
Reading your report, I believe one of the reasons our numbers differ is that you are simulating 10,000 allocations; my analysis only looked at the 5,121 allocations done by RIPE NCC in 2003-2006. Since the effects of HD ratio are progressive, the more allocations you simulate, the higher the relative increase in address space consumption becomes.
This experiment has been repeated 1,000 times in order to determine a stable average value for the relative increase in address consumption corresponding to a change in the address allocation policies from uniform 80% to an HD Ratio of 0.96, assuming constant demand for addresses.
To get a feeling of how stable your average is, could you indicate what the variation, the standard deviation in these 1,000 repeats is? i.e. did all 1000 give you a number close to 46% or were they spread out a lot?
A related consideration is that of the adoption of such a policy proposal by all 5 RIRs.
From http://www.arin.net/policy/proposals/proposal_archive.html I understand ARIN already abandoned two proposals to use HD ratio for IPv4 allocations (nrs. 2004-2 and 2003-10).
Regards,
-- Rene
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- =-= Rene Wilhelm RIPE Network Coordination Centre Email: wilhelm@ripe.net Amsterdam, the Netherlands Phone: +31 20 535 4417 Fax: +31 20 535 4445 =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- =-=
At 12:55 PM 23/02/2006, Tony Hain wrote:
Have either of you run the simulations with other HDR values? Would .97 make a significant difference?
Good question... Heres a table of the ratio of total address allocations using the RIPE NCC data, comparing the address consumption under the current 80% utilization criteria with a number of values for an HD Ratio criteria. I've included the mean standard deviation to give you a sense of the stability of these results. (Again, the technique here is to conduct an "experiment" consisting of 1000 separate simulations of a batch of 10,000 allocations, and the "result" is the ratio of the address space allocated under an HD Ratio framework, as compared to the same simulated end use populations being served under the current fixed 80% criteria) (fixed width font may help here) HD Ratio Ratio Mean Std Dev 0.98 1.04868 0.02285 0.97 1.25899 0.03363 0.96 1.45854 0.03371 0.95 1.63073 0.02848 0.94 1.78332 0.01859 regards, Geoff
HD Ratio Ratio Mean Std Dev 0.98 1.04868 0.02285 0.97 1.25899 0.03363 0.96 1.45854 0.03371 0.95 1.63073 0.02848 0.94 1.78332 0.01859
and what does .98 do to the flight ceiling of small folk? randy
At 02:07 PM 23/02/2006, Randy Bush wrote:
HD Ratio Ratio Mean Std Dev 0.98 1.04868 0.02285 0.97 1.25899 0.03363 0.96 1.45854 0.03371 0.95 1.63073 0.02848 0.94 1.78332 0.01859
and what does .98 do to the flight ceiling of small folk?
randy
I'll respond to this question, but in the interests of not wishing to overwhelming a whole swag of mailing lists I'll make this my last posting on this topic today. An HD Ratio of 0.98 imposes a higher efficiency target than the existing 80% rate for all prefix sizes smaller than a /16, and lower than 80% for allocations greater than a /16 (e.g. an HD Ratio of 0.98 implies an efficiency threshold of 72% for a /9 allocation.) As an example, if you had an end use population of between 3,277 and 6,554 numbered devices you would qualify for a /19 allocation under an 80% rule, while under an HD Ratio of 0.98 the end use population is between 3,468 and 6,841, corresponding to a required address efficiency level of 84% on this address block in order to qualify for a further address allocation. The use of an HD Ratio of 0.96 corresponds to an 80% efficiency level for a /24, so that 0.96 is no worse than 80% for all allocations, whereas HD Ratios greater than 0.96 impose an efficiency constraint greater than 80% on the smaller address blocks (/16 through to /24) - this can be easily modelled on any spreadsheet of course. regards, Geoff
Hi Rene, The results for these 1,000 experiments (with each experiment using 10,000 allocations as described in the original note) is a ratio of the address consumption using HD Ratio to that of a flat 80% is 1.4585421463303 with a Mean Standard Deviation = 0.0337090355213951 regards, Geoff At 12:43 PM 23/02/2006, Rene Wilhelm wrote:
This experiment has been repeated 1,000 times in order to determine a stable average value for the relative increase in address consumption corresponding to a change in the address allocation policies from uniform 80% to an HD Ratio of 0.96, assuming constant demand for addresses.
To get a feeling of how stable your average is, could you indicate what the variation, the standard deviation in these 1,000 repeats is? i.e. did all 1000 give you a number close to 46% or were they spread out a lot?
participants (4)
-
Geoff Huston -
Randy Bush -
Rene Wilhelm -
Tony Hain