BGP Global Statistics

What it does

BGP Global Statistics puts a single network in the context of the whole routing system. The top cards and the transit-provider ranking are a reference snapshot maintained by the site, covering the approximate size of the IPv4 and IPv6 tables, the number of active ASNs, year-over-year growth and average AS-path length. The ASN lookup is live: it queries RIPEstat for the AS holder name and counts the IPv4 and IPv6 prefixes the AS currently originates.

Technical background

The default-free zone is the set of routers that carry a full Internet routing table and no default route. Its size has grown almost every year since BGP-4 (today specified in RFC 4271) became the Internet's interdomain protocol. In August 2014 the IPv4 table crossed 512,000 routes, the default size of the hardware forwarding tables on several widely deployed router line cards, and some networks suffered outages until they reallocated TCAM space; operators still refer to it as 512k day. The IPv4 table has since grown toward one million routes, and the IPv6 table has passed two hundred thousand, with the relative growth rate of IPv6 well above that of IPv4.

Two forces drive IPv4 growth after exhaustion. First, the transfer market fragments space into small blocks, and the /24 is the smallest unit that can be routed globally, so each sold or leased /24 tends to become a separate route. Second, networks deaggregate for traffic engineering and for protection against more-specific hijacks. Both are rational for the individual network and costly for the system. The CIDR Report and similar long-running analyses list, for every AS, how many routes it could save by aggregating, and the same networks often appear near the top of those lists for years.

The number of ASNs in the table, in the high tens of thousands, tells a similar story of decentralization: every enterprise, university and regional ISP that multihomes adds one. The total assigned by the RIRs is considerably higher because many registered ASNs are not announced. Meanwhile the average AS-path length has stayed around four hops and drifted downward as large content and cloud networks peer directly with access networks at exchanges, flattening the old hierarchy of tier-1 carriers at the top.

The practical consequences are concrete. A router that takes full tables needs FIB space for both families plus headroom, and its control plane needs memory for the RIB and for every additional full-feed session. Convergence after a session reset scales with table size. Many enterprises therefore accept only default routes or partial tables from their providers.

For a single network, a prefix count reveals strategy: an ISP announcing hundreds of IPv4 routes and one IPv6 /32 probably carries legacy or acquired space and deaggregates for traffic engineering. Comparing the live counts for your ASN with networks of similar size and region is a quick way to see whether your announcements are proportionate.

Common errors and how to fix them

Problem Buying edge routers sized for today's table without headroom.

Fix Size FIB capacity for both the current IPv4 and IPv6 tables plus several years of growth, and check how the vendor partitions TCAM between families and other features such as ACLs.

Problem Taking multiple full tables on a small router that only needs to reach two providers.

Fix Accept a default route plus each provider's customer routes, or a partial table, and keep full tables only on routers that make per-destination decisions.

Problem Announcing every /24 of a large block when a few aggregates would do.

Fix Announce aggregates and add more-specifics only where traffic engineering or protection requires them; review your entry in aggregation reports periodically.

Problem Treating the reference figures on this page as exact current values in reports.

Fix Use them for scale and trends only, and cite a live table, a routing report or your own router's counts with a date when you need precise numbers.

Problem Ignoring growth in IPv6 route count because the absolute number is smaller.

Fix Plan IPv6 FIB space explicitly; on many platforms each IPv6 route consumes more hardware resources than an IPv4 route, so the effective share is larger than the raw count suggests.

Do it from the command line

# Holder and announcement status for an ASN
curl -s "https://stat.ripe.net/data/as-overview/data.json?resource=AS3356" | python3 -m json.tool | grep -E 'holder|announced'
# Count prefixes it originates
curl -s "https://stat.ripe.net/data/announced-prefixes/data.json?resource=AS3356" | grep -o '"prefix"' | wc -l

# Prefixes originated by an ASN, split by family
$p = (Invoke-RestMethod "https://stat.ripe.net/data/announced-prefixes/data.json?resource=AS6939").data.prefixes.prefix
"IPv4: $(($p | Where-Object { $_ -notmatch ':' }).Count)  IPv6: $(($p | Where-Object { $_ -match ':' }).Count)"

# Table size on a Linux router running BIRD
birdc show route count
# Same on FRRouting
vtysh -c 'show ip bgp summary' -c 'show bgp ipv6 unicast summary'
# Routes per family as seen by RIS for a country
curl -s "https://stat.ripe.net/data/country-routing-stats/data.json?resource=AR" | jq '.data.stats[-1]'

Frequently asked questions