CONCEPT Cited by 3 sources
Point of presence¶
A Point of Presence (PoP) is a physical deployment site — typically a data centre (or rack in a colocation facility, or an ISP edge cage) — where a network operator has servers, routers, and peering connections to adjacent networks. "PoP count" and "PoP density" are load-bearing specs in CDN / edge / DDoS literature because they govern how close a network's serving capacity sits to end users.
For the sysdesign-wiki, a PoP is:
- A latency primitive: the closer the PoP to the user, the lower the physical-RTT floor of any connection (see concepts/network-round-trip-cost, concepts/connection-time).
- A capacity primitive: under anycast, each PoP absorbs a geographic share of inbound traffic — including attack traffic (see systems/magic-transit) — so fleet scale = sum of per-PoP capacity.
- A reliability primitive: losing a single PoP degrades but doesn't kill an anycast service; losing PoPs globally (via BGP withdrawal — see concepts/bgp-route-withdrawal) does.
PoP density as a latency lever¶
Cloudflare's 2026-04-17 performance update (Source: sources/2026-04-17-cloudflare-agents-week-network-performance-update) cites three recent PoP deployments with measured RTT wins for the newly-nearest user population:
- Wroclaw, Poland — free-tier users 19 ms → 12 ms (−40 %) average RTT
- Malang, Indonesia — Enterprise traffic 39 ms → 37 ms (−5 %) average RTT
- Constantine, Algeria — new location (no RTT number published)
The headline framing: "Every millisecond shaved off a connection is a better experience for the real people using the applications and websites you build." PoP density is the infrastructure-axis lever on connection time, orthogonal to the software-axis levers (concepts/http-3, concepts/congestion-window).
But PoP density alone has diminishing returns on the ranking metric: Cloudflare explicitly says "adding new locations alone doesn't fully explain how we went from being #1 in 40 % of networks to #1 in 60 % of networks" — the marginal new PoP wins a small geography; the networks where a competitor was barely faster flip from software improvements that apply globally.
PoP as a peering / transit story¶
A PoP's useful-latency contribution depends on who the PoP peers with locally — not just where it's physically located. The wiki's anycast page notes:
"'Topologically nearest' is not 'geographically nearest'; peering / transit policy can route a client thousands of km out of the way. Anycast-CDN performance is partly a peering-engineering problem, not just a POP-density problem."
Putting a PoP in Jakarta doesn't help an Indonesian user unless that PoP has a local Internet-exchange peering session with the user's ISP. PoP deployment and peering engineering run as a joint programme.
PoP density and DDoS defence¶
Under anycast routing, a flood attack's aggregate bandwidth is automatically distributed across PoPs proportional to the attacker's geographic distribution. More PoPs → the same attack is spread across more per-PoP link capacity. The 7.3 Tbps Magic Transit attack (2025-06-20) was "detected and mitigated in 477 data centers across 293 locations" (Source: sources/2025-06-20-cloudflare-how-cloudflare-blocked-a-monumental-7-3-tbps-ddos-attack).
PoP density and reliability¶
Losing one PoP is a region-level degradation; every other PoP continues to serve. Losing a BGP advertisement from every PoP collapses the service globally — the 1.1.1.1 2025-07-14 incident (Source: sources/2025-07-16-cloudflare-1111-incident-on-july-14-2025) is the canonical wiki instance of an anycast service going globally dark in seconds because every PoP stopped advertising the prefix. PoPs are a reach multiplier in both directions.
Scale reference¶
- Cloudflare (2025-2026 posts): ~330+ cities globally, with active deployment into new geographies; the 7.3 Tbps writeup hit 477 datacenters / 293 locations, so the fleet is in that ballpark.
- Amazon CloudFront: ~600+ PoPs; bigger geographic spread, different internal design.
- Akamai: ~4,100+ PoPs; historically the densest CDN footprint, embedded deep inside ISPs.
Different CDNs optimise the PoP-density curve differently — Akamai's "inside every ISP" model trades ops complexity for latency, Cloudflare's "fewer, larger PoPs with dense peering" model trades PoP count for consistent software rollout.
Wiki framing¶
PoP is the physical substrate of the edge-network paradigm. Most of the wiki's edge-network concepts live on top of it: anycast is "advertise from every PoP"; service topology is "constrain to these PoPs"; densification is "add more PoPs where users are"; hot-path work is "make per-request code in each PoP faster".
Seen in¶
- sources/2026-04-17-cloudflare-agents-week-network-performance-update — canonical wiki instance of PoP deployment numbers (Wroclaw 19 → 12 ms / Malang 39 → 37 ms / Constantine) as one axis of the Sept → Dec 2025 ranking shift.
- sources/2025-06-20-cloudflare-how-cloudflare-blocked-a-monumental-7-3-tbps-ddos-attack — PoP count as DDoS-defence primitive: 477 data centres / 293 locations each absorbing a geographic share of the 7.3 Tbps flood.
- sources/2025-07-16-cloudflare-1111-incident-on-july-14-2025 — inverse instance: PoP-wide BGP withdrawal of the 1.1.1.1 Resolver prefixes collapsed the service globally at Internet-routing speed.