Mysticeti performance and validator risks

Mysticeti reduces latency by 80% compared to Bullshark and moves it from 1.9s to 400ms. It also lowers consensus CPU usage in production by 40%. This protocol operates on an uncertified DAG where validators sign and share blocks without needing a supermajority of 2f+1 signatures for certification. A block at round R commits if 2f+1 blocks in round R+1 reference it and those R+1 blocks are themselves referenced by 2f+1 blocks in round R+2. I find the reduction in round-trip overhead to be the strongest part of this protocol. Mysticeti uses proposer slots to classify blocks as to-commit, to-skip, or undecided. If the direct decision rule fails, validators use the indirect decision rule by searching for an anchor. An anchor is the lowest slot with round R’ > R that is already decided. If an anchor is marked to-skip, the validator looks for the next anchor. When they find the first anchor marked to-commit, the slot inherits that decision if 2f+1 blocks link to it. For transactions involving owned objects, the Mysticeti-FPC fast path provides 0.25 second latency under low loads. This fast path achieves 175k TPS at <0.5s latency with 10 nodes, which is 8x to 10x higher than existing fast path protocols. The protocol also uses a fast path for single-owner object transactions. This allows a validator to include a transaction for which other validators then vote through the Mysticeti DAG without additional signatures. This process enables a transaction to be certified without a mass of signature generation and verification, so it allows five to 10 times faster path transactions to be processed by the same CPU. In the bad case where a candidate proposer slot has a crashed leader, the protocol uses a novel commit rule to detect benign crash faults after 2 rounds. This prevents head-of-line blocking in the pipeline.

Validator economics and hardware requirements

Sui uses a delegated proof-of-stake system where the total voting power stays at 10,000. The network caps the voting power of an individual validator at 1,000, which is 10% of the total 10,000 voting power available during any given epoch, so no single entity can control more than a tenth of the network. To be an active validator, you must satisfy a minimum staking requirement of 30 million SUI. If your stake falls below 15 million SUI, the network removes you at the next epoch boundary. The SIP-39 proposal intends to lower this entry barrier by requiring 3 voting power instead of a specific SUI amount. You know how much these hardware costs weigh on small operators. Validators earn an average of 2.34% APY on their stake, which includes yield, commissions, and an automatic storage-cost rebate.

Running a validator requires 128 GB of RAM and a 4 TB NVMe SSD with sustained high IOPS. A 1 Gbps dedicated internet connection is also necessary.

Network stability and outages

Reliability remains a problem. On May 28, 2026, a gas charging logic bug in version v1.72 caused a 6 hour 44 minute outage. This defect froze $1 billion in on-chain value, including SUI tokens, DeFi liquidity, and NFTs. This follows a 2 hour transaction scheduling bug in November 2024 and a 6 hour consensus divergence in January 2026. I observe that the downtime length increased from 2 hours to 6 hours to 6.7 hours over these three incidents. The May outage occurred during a scheduled epoch change when a latent bug in how validators preserve randomness state across restarts triggered the crash. This happened because a specific transaction structure conflicted with gas calculation processes in v1.72. This caused all validators to crash and enter a loop. During that period, users experienced liquidation delays in protocols like Scallop and Cetus. SUI price dropped 8% to $0.91 following the incident. Will these technical failures prevent institutional users from committing to the network?