Abstract:High-performance flash-based key-value stores in data-centers utilize large amounts of DRAM to cache hot data. However, motivated by the high cost and power consumption of DRAM, server designs with lower DRAM per compute ratios are becoming popular. These low-cost servers enable scale-out services by reducing server workload densities. This results in improvements to overall service reliability, leading to a decrease in the total cost of ownership (TCO) for scalable workloads. Nevertheless, for key-value stores with large memory footprints these reduced DRAM servers degrade performance due to an increase in both IO utilization and data access latency. In this scenario a standard practice to improve performance for sharded databases is to reduce the number of shards per machine, which degrades the TCO benefits of reduced DRAM low-cost servers. In this work, we explore a practical solution to improve performance and reduce costs of key-value stores running on DRAM constrained servers by using Storage Class Memories (SCM). SCM in a DIMM form factor, although slower than DRAM, are sufficiently faster than flash when serving as a large extension to DRAM. In this paper, we use Intel (R) OptaneT PMem 100 Series SCMs (DCPMM) in AppDirect mode to extend the available memory of RocksDB, one of the largest key-value stores at Facebook. We first designed hybrid cache in RocksDB to harness both DRAM and SCM hierarchically. We then characterized the performance of the hybrid cache for 3 of the largest RocksDB use cases at Facebook (WhatsApp, Tectonic Metadata, and Laser). Our results demonstrate that we can achieve up to 80% improvement in throughput and 20% improvement in P95 latency over the existing small DRAM single-socket platform, while maintaining a 43-48% cost improvement over the large DRAM dual socket platform. To the best of our knowledge, this is the first study of the DCPMM platform in a commercial data center.

Impact of Memory Hierarchy on Memory Encryption Performance

Modeling and Benchmarking Computing-in-Memory for Design Space Exploration.

FeFET-Based In-Memory Hyperdimensional Encoding Design

SecDDR: Enabling Low-Cost Secure Memories by Protecting the DDR Interface

Architecting Non-Volatile Main Memory to Guard Against Persistence-based Attacks

In-Memory Computing Architecture for Efficient Hardware Security

Security Memory System for Mobile Device or Computer Against Memory Attacks

Protecting In-memory Data Cache with Secure Enclaves in Untrusted Cloud.

Embedding Security into Ferroelectric FET Array via In-Situ Memory Operation

Improving Performance of Flash Based Key-Value Stores Using Storage Class Memory As a Volatile Memory Extension

Efficient and Lightweight In-memory Computing Architecture for Hardware Security

From Ideal to Practice: Data Encryption in eADR-based Secure Non-Volatile Memory Systems

HPME: A High-Performance Hardware Memory Encryption Engine Based on RISC-V TEE

Embedding security into ferroelectric FET array via in situ memory operation

Configurable in-memory computing architecture based on dual-port SRAM

Efficient In-Memory AES Encryption Implementation Using a General Memristive Logic: Surmounting the Data Movement Bottleneck.

Fault Attacks on Encrypted General Purpose Compute Platforms

Low-latency DAE: A Configurable Lightweight Hybrid Data and Address Encryption Engine for IoT Real-time NVM Protection

An efficient run-time encryption scheme for non-volatile main memory

Research on Performance Optimization of Encryption Algorithms for Network Security Framework

A Reconfigurable Permutation Based Address Encryption Architecture for Memory Security