This is part 2 of a two-part series on Cloudflare's Q1 2021 DDoS data. Part 1 covers Highlights, Application-layer DDoS attacks, DDoS attack activity per industry, per source country, per target country, and ransom attacks.

Application-layer DDoS attacks continued: attack activity by industry, source, and target

A diagram showing a server unable to respond to legitimate users due to a malicious DDoS attack

Rankings moved this quarter

The most attacked industry this quarter is Telecommunication, up from sixth place in Q4 2020. Consumer Services is second and Security and Investigations third. The rankings not only reveal which industries are targeted, but also help us understand where DDoS activity is concentrated.

Where HTTP attack traffic originates

In an HTTP attack, the source IP cannot be spoofed; a connection needs to be established. So by looking up the geolocation of the source IP, we can tell which country sends the attack traffic. High DDoS activity rates show that large botnets operate in a given country. China was first in both Q4 2020 and Q1 2021, with the US close behind.

Which targets are under pressure

We also look at which countries bear the brunt of attacks, sorting DDoS activity by the customer's billing country. China and the US occupy first and second place. Last quarter India took first place from China, possibly because of the country's 2020 elections, which took place during Q4 2020.

Non-Enterprise plans face the most attacks and extortion

Customers on non-Enterprise plans are the most targeted by DDoS attacks, both by volume and by ransom DDoS (RDDoS) attempts. In Q1 2021, 13% of surveyed Cloudflare customers hit by a DDoS attack either received an extortion demand in an RDDoS attack or were threatened in advance. Most of them (62%) are on the Pro plan, and 33% are on the Business plan. This continues a trend from Q4 2020: back then, 17% of customers reported being extorted. That figure included a Fortune Global 500 company that we onboarded and protected after being targeted by a group claiming to be the Lazarus Group.

The RDDoS data comes from a survey of Cloudflare customers who reported being attacked. Because respondents are self-selected and report their own experiences, treat the figures as the share among those who were hit, not as a rate across all DDoS attacks.

What this part covered

In this part, we examined application-layer attacks through the lens of industry rankings, attack sources, target countries, and ransom attacks. In turn we saw that industry rankings shift quickly, that botnet geolocation can't be spoofed for HTTP attacks, and that extortion is pushing toward lower-tier plans. Taken together, these observations show where DDoS protection matters for application-layer traffic, and they point toward the shift in attacker behavior between Q4 2020 and Q1 2021.

Network-layer attacks: volume, size and timing

Where application-layer attacks target the Layer 7 service that end users reach, network-layer attacks go after exposed infrastructure itself — in-line routers and other network servers — and the Internet link they sit behind.

When network-layer attacks landed

Attack volume was uneven across the quarter. January was the busiest month for network-layer attacks, accounting for 42% of the quarter's total, followed by March at 34.2% and February at 23.8%.

BLOG-428 Embedded Image - ijyzK1

The largest attacks of the quarter, however, came in February, peaking at 300–400 Gbps.

BLOG-428 Embedded Image - ZlLDil

Measuring attack size

A Layer 3/4 attack can be sized two ways: bit rate (Gbps), the volume of traffic it delivers, and packet rate (pps), the number of packets delivered. High-bit-rate attacks try to saturate the Internet link. High-packet-rate attacks try to overwhelm routers and other in-line hardware.

For 2021 Q1, more than 97% of observed L3/4 attacks stayed below both 1 mpps and 500 Mbps — a continuation of the trend seen throughout the previous year. That does not make them harmless. Many organizations buy uplinks from their service providers with less than 1 Gbps of capacity, and the same public-facing interface usually carries legitimate traffic. An attack under 500 Mbps can therefore take down an unprotected property, and without cloud-based DDoS protection the disruption can be substantial.

BLOG-428 Embedded Image - McVAiY
BLOG-428 Embedded Image - TlXBtU

Short attacks as probes

More than 90% of attacks lasted under an hour. Short bursts may be aimed at causing damage without tripping detection systems, and manual-analysis mitigation can be useless against them — the attack finishes before an analyst can identify the traffic.

BLOG-428 Embedded Image - c7oRBE

Short attacks also work as reconnaissance. Load-testing and automated DDoS tools, widely sold on the dark web, can fire off a brief SYN flood and then follow up with a short attack using a different vector. This lets attackers gauge a target's defenses before deciding whether to escalate to larger rates and longer durations.

Attack vectors

SYN attacks remained the most popular vector in 2021 Q1, followed by RST, UDP and DNS amplification attacks.

A SYN flood exploits the foundation of a TCP connection. A stateful session begins with a three-way handshake: the client sends a packet with the SYN flag, the server replies with SYN-ACK, and the client confirms with an ACK, after which data can flow until the connection closes. Repeating SYN packets saturates the server or router's connection table, which tracks session state. The router answers with SYN-ACK, reserves memory for each connection and then sits waiting for an ACK that never arrives. Once enough of its memory is tied up, the router cannot serve legitimate clients — it crashes or stops accepting connections, and a denial of service results.

RST amplification flood attacks work along similar lines, draining the target servers' resources for looking up incoming packets against a current session.

Emerging threat vectors

BLOG-428 Embedded Image - 4FA0Xw

SYN remained dominant, but attacks aimed at Jenkins servers rose 940% for the quarter. Jenkins, a free open-source automation server used to support software development, left a flaw in older versions (Jenkins 2.218 and earlier) that helped attackers launch DDoS attacks. The issue was resolved in Jenkins 2.219 by disabling UDP multicast/broadcast messages by default, yet many vulnerable, exposed devices still run UDP-based services that are being harnessed for volumetric amplification attacks.

Cloudflare also observed a 433% increase in L3/4 attacks over QUIC, an encrypted-by-default Internet transport protocol running over UDP. Servers send Version Negotiation packets to tell clients which QUIC version they support; because UDP is stateless, an attacker can spoof the source address, mimic those packets and flood a client. On Cloudflare the traffic may have been meant to affect Cloudflare's own infrastructure — for instance by pushing versions downward — rather than specific customers.

TeamSpeak, the proprietary UDP voice-over-IP service gamers use to talk in real time, was the third emerging vector, up 203% quarter over quarter. Targeting a TeamSpeak server can cut a rival squad's real-time communications and hurt its performance in multiplayer games; talking rather than typing is a real advantage, which makes the channel a target.

A geographical caveat

Ranked by country, Rwanda, China and Brunei saw the most L3/4 attacks — but the ranking is not a measure of where attackers sit. In network-layer attacks, source IPs are routinely spoofed to hide origin. When analyzing L3/4 traffic, Cloudflare buckets it by the edge data center where it was ingested rather than by source-IP location, which sidesteps spoofing and still yields accurate geography given data centers in more than 200 cities.

Region and country breakdowns are available on the Radar DDoS Report dashboard's interactive map.