Source linked

DiStash Puts DRAM, SSD, and NVM Under One Transactional Roof

arxiv.org@systems_wire2 hours ago·Systems Engineering·2 comments

DiStash, a disaggregated transactional key-value store built on FoundationDB, runs a single atomic transaction across DRAM, SSD, HDD, and NVM pools-eliminating race conditions and tiering complexity.

distashfoundationdbebaytransactional key value storedisaggregated storagesystems engineering

A single transaction that atomically reads from DRAM, writes to SSD, and replicates to NVM—without custom application logic—is exactly what DiStash delivers.

DiStash is a disaggregated transactional key-value store that treats each storage medium (DRAM, SSD, HDD, NVM) as a separate stash pool. Applications issue one transaction to manipulate copies of key-value pairs across any combination of those pools. No more hand-rolled coordination or distributed locking between tiers.

One Transaction to Rule All Storage Tiers

Stash pools are configured as either ephemeral or durable. The application decides whether pools are inclusive (replicated) or exclusive (tiered). DiStash handles the transactional guarantees across those decisions. That means you can keep a hot copy in DRAM, a warm copy on SSD, and a cold copy on HDD—all updated or rolled back together.

Race conditions vanish. When two clients try to update the same key across different stash pools, DiStash serializes the transaction. Partial failures don't leave your data inconsistent across tiers. This is the kind of semantics you expect from a single-node database, now stretched across physically disaggregated storage.

Built on FoundationDB, Tested on eBay's Workload

The team behind DiStash—from eBay and USC—implemented it by extending FoundationDB. They quantified design tradeoffs with microbenchmarks and a production workload from eBay. The code is open source at github.com/ebay-USC/DiStash, so you can kick the tires yourself.

FoundationDB gives them a proven distributed transaction layer. DiStash adds the stash abstraction on top: each stash pool is a separate storage backend, but the transaction layer sees them as one logical store. The result is a system that lets you mix media without writing custom consistency code.

If you're building a multi-tier storage system or a disaggregated architecture, DiStash shows the pattern works—one transaction, any pool, no race conditions.

Source: DiStash: A Disaggregated Multi-Stash Transactional Key-Value Store
Domain: arxiv.org

Read original source ->

External source stays available while the OJO article and comment thread stay local.

More in Systems Engineering

view topic
Random Forest Surrogate Predicts HPC Performance from Just 54 Benchmarks

A cold-start random forest trained on 54 LAMMPS+SPICA runs ranks hybrid MPI+OpenMP configurations with 4.0% relative error - but only within the same hardware regime.

97-99% Accuracy Retained While Slashing TPOT: Cascaded LLM Serving That Routes Smart

A two-stage cascaded framework keeps 97-99% of the strongest model's accuracy but directs only hard queries to expensive models-an interpretable hyperparameter controls the cost-accuracy knob.

Host-Driven SRv6 Flowlet Balancing Cuts Tail Latency 33% vs ECMP

Shifting flowlet detection from switches to hosts and using SRv6 for stateless steering reduces tail latency by 33% over ECMP and 15% over random flowlet balancing in testbed experiments.

5G Network Jitter Slows Power Grid Optimization - Adaptive Thresholds Cut Convergence by 26%

Real 5G latency variability degrades ADMM-based distributed optimal power flow; an adaptive delay threshold mechanism on Raspberry Pi controllers reduces convergence time by 26.42% compared to static optima.

Comments load interactively on the live page.