Internet expansion, refinement, and churn
Archived MagicPoint presentation slides, compiled into a single PDF document.
2002_nanog0202.pdf (48 slides, 5.2 MB)
Slide text transcript
Slide 1: Internet expansion
Internet expansion refinement and churn Andre Broido broido@caida.org, CAIDA/SDSC/UCSD Evi Nemeth evi@caida.org, CAIDA and U.Colorado kc claffy kc@caida.org, CAIDA/SDSC/UCSD 12 feb 02
Slide 2: Overview
Overview Data description and caveats Taxonomy of prefixes (definitions) Size measures & proportions Topology changes Prefix set changes More specifics Multihoming Routing updates Acknowledgement: inspired by Jeff Huston's analysis
Slide 3: Data analysis: BGP peer and prefix selection
Data analysis: BGP peer and prefix selection Select only full-size tables from RouteViews peers full: > 104K prefixes filtered: 89-97K prefixes Prefix selection: globally routed prefixes (those common to all tables) semiglobal (found in the majority of backbone tables) prefixes dropped by one (few) peers still observed prefixes seen only locally won't get into table more robust than either full or global prefix set We use only semiglobal prefixes
Slide 4: Caveats:
Caveats: many metrics are sensitive to prefix/peer choice important when trends are small and/or undecided instant and daily influx or loss of prefixes Table sizes differ noticeably (few %) by day
Slide 5: Data sources
Data sources UOregon ANTC RouteViews BGP tables Data on or around the 1st of each month Nov. 1, 2000 (18 tables) Nov. 1, 2001 (26 tables) Feb. 1, 2002 (26 tables) unless otherwise specified Sampling for prefix analysis: from 1 year (1997-2001) to 2 hours (data from the end of 2001.)
Slide 6: Prefix taxonomy
Prefix taxonomy Prefix types: standalone -- no subsets/supersets in the table root -- a least specific route, has subsets in the table more specific -- a subset of some other prefix In addition: top prefixes = standalone + roots Caveats: depends on the prefix set not an inherent property of a prefix
Slide 7: Bulk and relative measures
Bulk and relative measures Short term (3 month) trends, Feb.2002 Semiglobal prefixes: 104K, stable Prefix table size is constant for 4 months (-1k) Smallest blocks (/24s): 55.8%, down More specifics: 50.3%, down Trends of 1990s reverted ASes: 12570, up Transit ASes: 2050, up Transit ASes: 16%, stable
Slide 8: Bulk and relative measures (cont'd)
Bulk and relative measures (cont'd) AS links in BGP graph: 26046, up AS links/ASes 2.06, stable/down AS paths/ASes (1 tb.) 1.25, stable 4 years Path system very rigid Less paths than the graph allows Due to policy constraints & shortest paths
Slide 9: Bulk and relative measures (cont'd)
Bulk and relative measures (cont'd) ASes with one prefix: 40% Prefixes originated by such ASes: 5% very small contribution ASes with over 100 prefixes: 1% <---"large ASes" Prefixes originated by such ASes: 32%
Slide 10: Growth in prefixes, ASes, and degree
Growth in prefixes, ASes, and degree 1999 2000 2001 Semiglobal prefixes 64769 88714 (+37%)102394 (+15%) ASes 6107 9116 (+50%) 12399 (+36%) AS links/nodes (avg in/outdegree) 1.97 2.00 2.07 prefix and AS growth slowed in 2001 prefix growth is close to 2/3 of AS growth link/node ratio grows slowly, almost invariant peering richness, and navigation complexity of BGP AS graph did not significantly change between Nov 2000 and May 2001, although there were a lot of changes for individual ASes
Slide 11: Is average AS path length decreasing?
Is average AS path length decreasing? Since 1999, many AS paths changed either way Change in the average AS path length is insignificant
Slide 12: Peering richness: changes per AS, not on average
Peering richness: changes per AS, not on average
Slide 13: AS refinement formula: P = 200 A^(2/3)
AS refinement formula: P = 200 A^(2/3)
Slide 14: Super-refinement
Super-refinement Refinement formula worked for 3 years Current prediction: 107140 (formula overestimates) Prefix growth stopped (last few months) AS refinement faster than predicted by formula Average prefixes per AS: 8.3, down 4 years Will eventually have less than one prefix per AS Already have ASes with 0 announced prefixes (85 transit-only ASes for Nov 01, 2001, up from 55 a few months ago)
Slide 15: Prefix distribution: findings
Prefix distribution: findings
Changes in global routing not captured by bulk
measures such as number of prefixes in a table
Net change is a sum total of commensurable
contributions of opposite sign, so...
Total variation is not the difference,
but rather the sum of their magnitudes.
Most measures of routing system complexity
demonstrate slow growth, dynamic equilibrium
and/or decrease for May-Dec. 2001.
Current state of the Internet routing system
shows large variation but NOT rapid growth!
Slide 16: Prefix distribution
Prefix distribution
/24 prefixes are the smallest globally routable
Q: Is proportion of /24s in backbone tables growing?
A: No
Since 1998, the share of /24s has stayed between 57% and 58.5%
Has decreased since then by 0.6%-0.9% in all non-filtered tables
About 1/2 of all prefixes are more specifics (subsets).
1999 50%
2000 55%
2001 52%
2002 50.3%
Slide 17: Prefix distribution: refinement of prefix trees
Prefix distribution: refinement of prefix trees Prefix type breakdown (semiglobal prefixes in 26 tables, Nov.01,2001) standalone 44264 42.75% root 5601 5.41% more specifics 53686 51.84% total prefixes 103551 100.00% Percentage of root prefixes grows slowly 1999 4.64 2000 4.86 2001 5.41 Another refinement phenomenon: growing share of the table is roots constant or decreasing share of more specifics average number of subblocks of a root is decreasing
Slide 18: Prefix set churn
Prefix set churn Churn (emergence and disappearance of prefixes in the table) was much higher than growth rate. highest for more specifics lowest for standalone blocks net change in table size sometimes < 0 semiglobal prefixes in following data uses all available RouteViews backbone tables, including filtered tables, for 3 sample points: May 3, August 1, and November 1, 2001 (33, 42, and 39 tables, respectively)
Slide 19: Changes in prefix status from May->Aug 2001
Changes in prefix status from May->Aug 2001
transitions are from columns to rows
from May to August 2001
equal number of prefixes moved from standalones to roots/m.s.'s
more prefixes changed from m.s. to standalone than v.v.
"Vacuum" counts prefixes which appeared or disappeared
St.al Root M.sp Vac Sum Out
St.al 38676 544 576 2450 42246 8.45
Root 377 4374 40 307 5098 14.20
M.sp 917 41 42188 8746 51892 18.70
Vac 4232 420 9206 0 13858 100.00
In 13.08 19.71 18.93 83.01 99236r
Sum 44202 5379 52010 11503 101591c
Slide 20: Changes in prefix status from Aug->Nov 2001
Changes in prefix status from Aug->Nov 2001
transitions are from columns to rows
from August to November 2001
Root prefixes coming mostly from standalone (607 vs. 433 new)
more prefixes changed from standalone to m.s. than v.v. (reversal of May-Aug trend)
standal. root more spec vac sum out
St.al 39530 607 1460 2605 44202 10.57
Root 396 4532 146 305 5379 15.75
M.sp 604 33 41835 9538 52010 19.56
Vac 3738 433 10178 0 14349 100.00
In 10.72 19.95 22.66 86.75 101591r
Sum 44268 5605 53619 12448 103492c
Slide 21: Prefix set churn, cont.
Prefix set churn, cont.
May-Aug 2001 (33 (May) and 42 (Aug) RouteViews peers):
% standalone % of root % more spec.
Out 8.45 14.20 18.70
In 13.08 19.71 18.93
Net 4.63 5.51 0.23
No growth for more specifics in May-Aug 2001
Aug-Nov 2001 (42 (Aug) and 39 (Nov) RouteViews peers):
% standalone % of root % more spec.
Out 10.57 15.75 19.56
In 10.72 19.95 22.66
Net 0.15 4.20 3.09
No growth for standalones in Aug-Nov 2001
("In" and "Out" include transitions when existing prefix
changes its type, e.g., from standalone to more specific)
Slide 22: Growth of non-transit ASes
Growth of non-transit ASes
Transit and non-transit ASes contribute about equal numbers of prefixes despite the fact that there are 5 times as many non-transit ASes as transit ASes.
Transit ASes announce over 1/2 prefixes (53%)
For 1999-2000-2001:
Fraction of nontransit ASes: 81%, 83%, 84%
Announcements from nontransit ASes
(percentage of semiglobal): 42%, 43%, 46%
Slide 23: Growth of non-transit ASes, cont.
Growth of non-transit ASes, cont.
Q: Does the Internet grow mostly at the periphery?
A: Most likely, although BGP graph may not have
enough coverage at network edge to answer
may show transit ASes as nontransit
may undercount indegree (multi-homing)
captures only downstream, not lateral, connectivity
18%, 17%, 16% transit ASes in 1999, 2000, 2001
absolute number grows but proportion shrinks
bidirectional part of the graph (core): 3.13%, 2.24%, 2.09%
compare to skitter (traceroute-based) AS graph:
32% transit ASes, twice as many as BGP graph
28% ASes in combinatorial core (BGP: 2%)
Slide 24: Multihomed ASes
Multihomed ASes
Definitions: AS with indegree > 1 in BGP AS graph is called multihomed
AS with outdegree > 0 in BGP AS graph is called transit
BGP gives only a lower bound on multi-homing,
although a consistent one
(26 vs. 39 tables differ by 0.3% in multi-homed ASes)
For Nov 01, 2001 RouteViews data, multihomed networks originated
76% of all prefixes
74% of all more specific (subset) prefixes
includes transit & non-transit
multihomed nets originate a proper fraction of more specifics
Non-transit multihomed ASes contribute less prefixes than their fair share:
Year %AS %prefixes
1999 43.65 27.46
2000 45.84 29.37
2001 48.75 29.74
Fraction of non-transit multihomed ASes grew in 1999-2001
Fraction of prefixes grew in 1999-2000, was constant in 2000-2001
Slide 25: Transit AS prefixes by type (Nov.01 2001, 26 tables)
Transit AS prefixes by type (Nov.01 2001, 26 tables) Prefixes from transit multihomed ASes: standalone 23801 49.28% root 3723 7.71% more specific 20777 43.02% total 48301 100.00% --> 46.64% of all prefixes Prefixes from transit singlehomed ASes: standalone 2608 37.74% root 344 4.98% more specific 3958 57.28% total 6910 100.00% --> 6.67% of all prefixes
Slide 26: Non-transit AS prefixes by type
Non-transit AS prefixes by type Prefixes from non-transit multihomed ASes: standalone 10715 34.80% root 999 3.24% more specifics 19080 61.96% total 30794 100.00% --> 29.74% of all prefixes Prefixes from non-transit singlehomed ASes: standalone 6507 39.51% root 452 2.74% more specifics 9509 57.74% total 16468 100.00% --> 15.90% of all prefixes
Slide 27: New top prefixes
New top prefixes New address space announcements and deaggregation of existing prefixes were two major sources of new root and standalone prefixes between Nov 2000-May 2001.
Slide 28: Multihomed ASes and traffic engineering
Multihomed ASes and traffic engineering Q: Are extra routes introduced for traffic engineering? A: Yes, but only a minority Multihomed ASes originating on one link Nov.01, 2001, 26 tables; multiorigin prefixes are excluded 22.2% of all prefixes 15.8% of all top prefixes 28.6% of all more specifics more specifics are announced on one link more frequently 6-12% more specifics may be traffic engineered splinter blocks Majority of semiglobal prefixes are announced on all available connections Largest groups: On 1 connection for singly homed AS, 22.6% On 2 connections for doubly homed AS, 15.8% On 3 connections for triply homed AS, 31.1% On 2 connections for triply homed AS, 7.3% (triply homed = indegree >= 3)
Slide 29: More specifics and traffic engineering
More specifics and traffic engineering traffic engineering would imply the more specific announced on only one connection some more specifics are involved in traffic engineering, but most of them aren't
Slide 30: Routing changes: Oct -Nov 2001
Routing changes: Oct -Nov 2001 Semiglobal prefixes, 26 full size backbone tables 104555 prefixes on Oct 01 103815 prefixes on Nov 30 Union: 137374 There are lots of flips (withdrawals and reannouncements) transients (initial announcements and final withdrawals)
Slide 31: Flips (reannouncements after withdrawals)
Flips (reannouncements after withdrawals)
number % of total % of Oct01
Long-lived: 93127 67.79% 89.07%
Apparently emerged: 32819 23.89% 31.39%
Apparently disapp.: 33559 24.43% 32.10%
Transients (came&went): 22131 16.11% 21.17%
Entries w.a flip: 31007 22.57%
Long-lived entries w.a flip: 20951 15.25%
Long-lived entries' flips: 35591 60.06% (of flips)
Transient entries w.a flip: 3743 2.72%
Transient entries' flips: 9285 15.67% (of flips)
long-lived + emerged + disappeared - transients = total observed = 137K
Slide 32: Flips (/24's flip more frequently; /21-24's most transients)
Flips (/24's flip more frequently; /21-24's most transients)
Pf.len prefxs flp.pf flips trans %all %flp.pfs %flips %trans
1 1 0 0 1 0.00 0.00 0.00 0.00
2 1 0 0 1 0.00 0.00 0.00 0.00
8 28 8 264 11 0.02 0.03 0.45 0.05
9 5 0 0 0 0.00 0.00 0.00 0.00
10 9 0 0 1 0.01 0.00 0.00 0.00
11 12 1 2 0 0.01 0.00 0.00 0.00
12 39 8 12 1 0.03 0.03 0.02 0.00
13 95 13 21 0 0.07 0.04 0.04 0.00
14 236 27 43 5 0.17 0.09 0.07 0.02
15 410 55 81 5 0.30 0.18 0.14 0.02
16 7387 1256 1966 115 5.38 4.05 3.32 0.52
17 1452 203 302 62 1.06 0.65 0.51 0.28
18 2605 401 581 115 1.90 1.29 0.98 0.52
19 7450 1213 1792 203 5.42 3.91 3.02 0.92
20 6852 1737 3981 356 4.99 5.60 6.72 1.61
21 7176 1193 1866 2009 5.22 3.85 3.15 9.08
22 8803 1792 3367 1169 6.41 5.78 5.68 5.28
23 11636 2600 4503 2028 8.47 8.39 7.60 9.16
24 83177 20500 40480 16049 60.55 66.11 68.31 72.52
Total 137374 31007 59261 22131 100.00 100.00 100.00 100.00
The fraction of flips in /24 range is larger than fair share
But not overwhelmingly large
Slide 33: Flips in /24s per class (A,B,C)
Flips in /24s per class (A,B,C)
Statistics of /24s
Class /24s flp.pfs flips %/24s %fpg/24s %allfp P{fp|cl}
A 11266 1890 3338 13.54 9.22 8.25 16.78
B 6666 2034 4587 8.01 9.92 11.33 30.51
C 65245 16576 32555 78.44 80.86 80.42 25.41
Total 83177 20500 40480 100.00 100.00 100.00 24.65
Slide 34: Flips per lifetime of a prefix
Flips per lifetime of a prefix #) AS Av.#pf Flips Fp/Kupt Fp/Klft Pc.fp Acc.pc NewAnn TmpW/drw Reann FinW/hdr 1) 4323 421.72 1466 4.89 4.59 2.47 2.47 34 52 51 16 2) 752 3.60 1444 564.72 335.58 2.44 4.91 1 659 658 2 3) 701 2196.94 1379 0.88 0.82 2.33 7.24 101 351 350 91 4) 702 956.48 924 1.36 1.32 1.56 8.80 68 373 346 45 5) 3908 652.29 921 1.99 1.79 1.55 10.35 63 105 109 43 6) 705 397.04 661 2.34 2.18 1.12 11.47 62 162 165 46 7) 1580 180.75 613 4.77 4.36 1.03 12.50 5 147 144 7 8) 3475 28.91 540 26.27 21.79 0.91 13.41 23 237 238 22 9) 4755 308.19 510 2.33 2.20 0.86 14.27 31 195 180 31 10) 8708 161.29 497 4.33 4.14 0.84 15.11 21 52 59 19 18) 8112 48.03 378 11.07 8.71 0.64 20.58 7 7 7 4 27) 724 244.60 278 1.60 1.56 0.47 25.14 12 76 81 11 39) 1239 692.89 190 0.39 0.38 0.32 30.21 55 94 90 33 56) 13609 109.98 150 1.92 1.90 0.25 35.03 14 40 40 4 80) 18618 64.45 114 2.49 2.45 0.19 40.17 1 10 9 2 110) 7629 21.04 83 5.55 5.40 0.14 45.10 7 22 23 6 150) 11623 6.90 66 13.46 13.26 0.11 50.01 0 12 12 0 201) 9800 104.11 52 0.70 0.70 0.09 55.07 6 18 17 2 266) 11554 18.78 39 2.92 2.90 0.07 60.01 0 3 3 1 356) 17625 6.84 29 5.97 4.97 0.05 65.05 4 7 9 3 399) 10223 42.14 25 0.83 0.83 0.04 67.01 1 10 10 2 478) 600 89.94 20 0.31 0.31 0.03 70.01 1 17 17 1 654) 6539 239.63 15 0.09 0.09 0.03 75.02 6 10 10 3 902) 224 14.94 10 0.94 0.94 0.02 80.00 0 10 10 1 1262) 2508 1.16 7 8.47 4.92 0.01 85.01 0 7 7 0 Total origins of flips: 5259 AS. Sum of temp. w/draw AS events: 22657. Reannouncement: 22816 All AS initiating temp. withrawals: 5259. Reannouncing AS: 5259 Note: when a prefix changes origin, an interval of AS lifetime sum is lost
Slide 35: Several origin ASes over time
Several origin ASes over time
#) #or.AS # Prefixes
1) 1 130786
2) 2 4901
3) 3 1500
4) 4 96
5) 5 50
6) 6 16
7) 7 17
8) 8 5
9) 9 1
10) 11 1
11) 12 1
Total 137374
6588 prefixes changed hands (origin or some of origins)
4220 prefixes were multiorigin (at least once)
3861 ( 91.49% of above) multiorigin prefixes also changed hands
multi-origin prefixes are highly dynamic, some may be
artifacts of convergence to a new origin at snapshot time
Slide 36: BGP table change contributors (summary)
BGP table change contributors (summary) ASes with new and/or (temporarily) lost prefixes sum total of all changes in prefix status 50% of (all) changes from 127 most "dynamic" ASes Sorted by total lost+found prefix occurrences for Oct-Nov 2001 (61 days, 2 hour int, 711 tables)
Slide 37: BGP table change: top 10 contributors
BGP table change: top 10 contributors
2-hour intervals with changes
type ASN Av.#pf ap+ds a+d a+d %a+d cum new tmp reapp final
/Kupt /Klft %a+d app disp disp
1 Bac 701 2196.94 6790 4.35 4.04 3.67 3.67 101 351 350 91
2 Bac 3908 652.29 5915 12.75 11.49 3.20 6.87 63 105 109 43
3 AOL 4323 421.72 3236 10.79 10.14 1.75 8.62 34 52 51 16
4 Bac 1239 692.89 3199 6.49 6.47 1.73 10.35 55 94 90 33
5 Res 752 3.60 2893 1131.4 672.32 1.56 11.92 1 659 658 2
6 Bac 702 956.48 2883 4.24 4.12 1.56 13.48 68 373 346 45
7 Tech 71 46.22 2589 78.79 77.60 1.40 14.88 3 10 10 6
8 Bac 705 397.04 2409 8.53 7.94 1.30 16.18 62 162 165 46
9 Bac 4999 21.64 1493 97.02 28.93 0.81 16.99 6 2 2 6
10 Mil 1580 180.75 1448 11.27 10.30 0.78 17.77 5 147 144 7
(ap = appeared, ds = disappeared)
Both background noise and storm-like events are present
Slide 38: BGP table change contributors: prefix flips (reannouncements after withdrawals)
BGP table change contributors: prefix flips (reannouncements after withdrawals)
50% flips from 150 ASes (3% of 5259 flipping ASes)
75% flips from 12.4% ASes
#events for AOL larger than for other ASes
not just background noise
Cum
type ASN Av.#pf Flips Fp/Kupt Fp/Klft %fps %fps newap tmpdisp re-app finaldisp
1 AOL 4323 421.72 1466 4.89 4.59 2.47 2.47 34 52 51 16
2 Res 752 3.60 1444 564.72 335.58 2.44 4.91 1 659 658 2
3 Bac 701 2196.94 1379 0.88 0.82 2.33 7.24 101 351 350 91
4 Bac 702 956.48 924 1.36 1.32 1.56 8.80 68 373 346 45
5 Bac 3908 652.29 921 1.99 1.79 1.55 10.35 63 105 109 43
6 Bac 705 397.04 661 2.34 2.18 1.12 11.47 62 162 165 46
7 Mil 1580 180.75 613 4.77 4.36 1.03 12.50 5 147 144 7
8 Mil 3475 28.91 540 26.27 21.79 0.91 13.41 23 237 238 22
9 Dev 4755 308.19 510 2.33 2.20 0.86 14.27 31 195 180 31
10 Dev 8708 161.29 497 4.33 4.14 0.84 15.11 21 52 59 19
Under best conditions, a prefix flips once in 3 months
Aside: global (all-peer) prefixes have twice as much flux as semiglobal (flux estimates depend on the prefix set)
Slide 39: analysis of BGP updates
analysis of BGP updates now we look at granularity of individual BGP updates RouteViews started archiving October 2001 we analyze BGP updates over one month (Nov 2001) 15 min updates per file, 22 peers contribute Contribution sizes span an order of magnitude use 50K cutoff to avoid events like full-table transmission leaves about 90% of all time intervals (2434 out of 2711) and about 2/3 of all packets Results: Announcements more frequent than withdrawals Many whole table transmissions, O(100K) prefixes Weekly and diurnal variations clearly visible Match traffic pattern on the 'Net Traffic load translates into path fluctuations Is it human intervention or automatic tools output?
Slide 40: BGP updates over one month
BGP updates over one month withdrawal vs announcemts (and some full table updates)
Slide 41: BGP prefix announcements
BGP prefix announcements ann/prefix related to counts of 15-min.int, esp. at the cusp
Slide 42: BGP prefix announcements/withdrawals
BGP prefix announcements/withdrawals Nov 2001 RouteViews 2711 files (no cutoff for #announcements per file) total packets: 30M packets with announcements 28.2M packets with withdrawals 1,955,188 packets with both 127,429 total announcements 83.5M total withdrawals 11.7M average announcements/packet 2.7736 average withdrawals/packet 0.3874 average withdrawals/wd.packet 5.9615 15X as many announcement as withdraws 15X as many withdraws as ann.&withdraw withdrawals packed at prefixes/AS path ratio of ~6.5
Slide 43: Dependence between packets seen & bgp announcements
Dependence between packets seen & bgp announcements some AS paths update 100s of prefixes at once
Slide 44: packing of prefixes into announcements
packing of prefixes into announcements two regions: full tables (blue) and incremental updates (green)
Slide 45: BGP updates contributors
BGP updates contributors Q: Are small ASes major contributors of BGP updates? A: The bulk of BGP announcements comes from ISPs who announce many (tens to hundreds) prefixes: backbone providers government networks ISPs in developing nations ISPs in Canada and US Large batches of announcements often contain deaggregated intervals of IP space (as in "avoiding a small block" scenario)
Slide 46: Top 10 announcers, Nov 2001
Top 10 announcers, Nov 2001 type ASN #Anns %ann cum% /15mi 1 ISP Can 852 777540 1.40 1.40 732 2 Dev Eu 12302 578932 1.05 2.45 681 3 Mil 6595 564201 1.02 3.47 1622 4 Bac 577 508675 0.92 4.39 1085 5 Dev Asia 4755 471458 0.85 5.24 1985 6 Bac 702 436282 0.79 6.02 2421 7 Dev SAm 20043 412334 0.74 6.77 749 8 Gov ZA 2018 370335 0.67 7.44 337 9 Gov 2572 365096 0.66 8.10 592 10 Bac 6453 346551 0.63 8.72 2442 top 2 announce 538 and 102 prefixes, resp. (backbones up to 2K) announcements are spread across time, not just storms reminder: only 1% of ASes announce >100 prefixes
Slide 47: Conclusions
Conclusions Internet undergoes refinement process Some bulk measures are stable Others change slowly Opposite trends cancel each other More churn than bulk measures suggest Internet churn comes from everywhere mostly from large ISPs, .gov/.mil, and developing nation telecoms Small ASes generate little churn and table growth Invariant or almost invariant relations do exist e.g. AS paths/AS ratio still a lot left to analyze...
Slide 48: www.caida.org/publications/presentations/
www.caida.org/publications/presentations/ Andre Broido kc claffy UCSD/SDSC/CAIDA kc@caida.org www.caida.org