Scalog
Scalog is a research shared log that replicates records within shards before assigning a total order across shards.
Records are replicated before receiving a global position. In the baseline protocol, componentwise minima of each shard's local reports identify fully replicated prefixes. Ordered cuts then determine a common record sequence; only afterward does the server acknowledge the append. [1]
An ordering-layer outage can leave local intake running while client commits wait. Finalize & Add requires clients to retransmit records outside the final cut; Remove & Replace pauses affected-shard writes; Mask spends extra replicas to preserve local service. [1]
Section 6.3.2 uses storage-server emulation beyond the 40-machine testbed. The 52 million records/s result additionally emulates aggregator layers and is an ordering-capacity experiment, not measured end-to-end durable-storage throughput. The associated 1.6 ms latency is estimated. [1]
The public reimplementation exposes append and read commands; a read takes a global sequence number and shard identifier. [2]
Relations
- related
- CORFU, Linearizability
Citations
- Ding et al. Scalog: Seamless Reconfiguration and Total Order in a Scalable Shared Log. NSDI 2020, pp. 325-338; sections 4.1, 4.4 and 6.3.2.
- Scalog repository README at cdc29ba5ca9acb11bb98dda995755c0d28ef939e.
Sources & caveats
- Historical design synthesis from the cited paper sections and pinned README; no hardware reproduction or current-product benchmark.
- The paper's baseline cut takes all replicas into account; the optional Mask recovery policy has a different replica requirement.
- The pinned repository README labels the code a reimplementation; this revision is not established as the artifact behind the paper's measurements. [2]