Internet traffic rose sharply in many countries during the first quarter of 2020 as populations sheltered in place, and network-layer DDoS activity rose with it. Cloudflare's quarterly observations show attackers shifting toward smaller, shorter floods, while a persistent minority continued to chase very large volumes.

Attack size: the long tail gets shorter

Most Q1 2020 attacks were small by bit rate. Ninety-two percent stayed under 10 Gbps, up from 84% in Q4 2019. The shift is sharper further down the distribution: 64% of network-layer attacks peaked below 500 Mbps in Q1, against 47% in the prior quarter.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - b7N4AD

Packet rates moved the same way, with the bulk of attacks peaking below 1 million packets per second. Taken together, the figures suggest attackers are putting less effort into generating high-rate floods, measured in bits or packets per second.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - wXn5Zj

Durations fell as well. Seventy-nine percent of Q1 attacks lasted between 30 and 60 minutes, compared with 60% in Q4 — a 19% increase.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - 2VWF8I

Two factors help explain the pattern. DDoS-for-hire services lower the barrier to entry: Kaspersky puts the cost of a 300-second attack at as little as $5, and free tools are enough for amateur actors to produce packet floods. Separately, an attack below 10 Gbps can still take down an underprotected property, and a short cheap burst may yield a strong return for an attacker seeking a ransom.

Large attacks persist, but are rarer

Big floods did not disappear. The largest network-layer attack Cloudflare observed and mitigated in the quarter arrived in March and peaked just above 550 Gbps.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - PoJyLC

Persistence and vectors

During the Q4 2019 holiday season, one Cloudflare IP absorbed as many as 523 attacks in a day, and each targeted IP saw an average of 4.6 attacks daily. In Q1 2020, as lockdowns took hold, the number of attacks climbed above the monthly average for the first time since that holiday period — but persistence fell. The average rate dropped to as low as 2.2 attacks per IP per day, with a maximum of 311 on a single IP, 40% below the holiday quarter.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - W2mQW2

The number of vectors used per IP per day held roughly steady across the two quarters at about 1.4, with a maximum of 10.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - CTwEms

More than 34 attack vector types appeared at L3/4. SYN attacks accounted for 60.1% of the total, ACK attacks 12.4% and CLDAP 5.3%; TCP-based SYN and ACK floods together made up 72% of L3/4 vectors.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - zApl6R

March: a surge in volume and size

Attack counts in March 2020 exceeded January and February, indicating attackers treated the crisis as an opening. The second half of March (March 16–31), after lockdown and shelter-in-place orders spread, saw 55% more attacks than the first half (March 1–15). Larger attacks also clustered late in the month: 94% of attacks peaking at 300–400 Gbps were launched in March.

Network-Layer DDoS Attack Trends for Q1 2020 Embedded Image - 4dCGei

Mitigation implications

Cloudflare argues that shrinking attack rates and durations make legacy 15-minute mitigation SLAs impractical, and says it mitigates network-layer DDoS attacks in under 10 seconds in most cases. Because more attacks are localized, scrubbing-center designs are limited by coverage and act as a choke point; Cloudflare's distributed architecture gives every one of its data centers across 200 cities full mitigation capability. For large volumetric events, the company cites a network capacity of over 35 Tbps across one of the most interconnected networks in the world, and expects an attack exceeding 1 Tbps in the future. In its framing, distributed capacity is what lets attacks large and small be stopped close to the source.