Transcription
Delivering NoSQL Database Performancewith NVMe SSD’sVijay BalakrishnanMemory Solutions Lab.SamsungFlash Memory Summit 2016Santa Clara, CA1
Agenda NVMe SSD Samsung PM1725 NVMe SSD Redis-On-Flash with PM1725 Deliver 1MOPS @ 1ms latency consistently PM1725 as NVMf target for Cassandra Build efficient remote storage for databasesFlash Memory Summit 2016Santa Clara, CA2
NVMe Design Advantages Lower latency Direct connection to CPU’s PCIe lanes Higher bandwidth Scales with number of PCIe lanes Best in class latency consistency Lower cycles/IO, fewer cmds, better queueing Lower system power- No HBA requiredFlash Memory Summit 2016Santa Clara, CA3
PM1725Samsung PM1725 Specification Leverages latest VNAND technology Delivers consistent low latencyFlash Memory Summit 2016Santa Clara, CAForm Factor2.5”Host InterfacePCIe Gen3 x4Capacities800GB, 1.6TB, 3.2TBSequential Read3300 MB/sSequential Write1900 MB/sRandom ReadUpto 840KIOPSRandom WriteUpto 130KIOPSRead Latency134 usecWrite Latency68 usec4
Redis-on-Flash Closed-source (RLEC Flash) 100% compatible with the open-source Redis Uses Flash as RAM extension Increases effective node capacity Tiering memory into “fast” and “slow”: RAM saves keys and hot values Flash saves cold values Dynamic configuration of RAM/Flash usage Uses RocksDB as the storage engine to optimizeaccess to block storage Multi-threaded and asynchronous Redisused to access FlashGet it Here Today: https://redislabs.com/rlec-flashFlash Memory Summit 2016Santa Clara, CA5
Why Redis-on-Flash? Optimize price-to-performance for a given workload DRAM is more performant than flash, but /GB is higher Tiering dramatically reduces /GB, while preserving goodperformance ( /ops) Limited DRAM capacity per serverEnables orders-of-magnitude more capacity per serverRoF is suitable for large datasets with skewed accessdistributionFlash Memory Summit 2016Santa Clara, CA6
System Under Test Single client, single server Industry-standard components, all available todayFlash Memory Summit 2016Santa Clara, CA7
Use case 1: 1KB Objects R/W:80/2050% RAM-to-Flash ratioFlash Memory Summit 2016Santa Clara, CA95% RAM-to-Flash ratio100% of requests served with 1msec latency8
Use case 2: 100B Objects R/W : 50/5050% RAM-to-Flash ratio85% RAM-to-Flash ratio100% of requests served with 1msec latencyFlash Memory Summit 2016Santa Clara, CA9
Comparison to SATA 80/20 read-write ratioFlash Memory Summit 2016Santa Clara, CA10
DRAM or Flash? Performance and lash Memory Summit 2016Santa Clara, CA11
NVMe Over Fabrics (NVMf) Reference: http://www.snia.org/sites/default/files/ESF/NVMe Under Hood 12 15 Final2.pdfFlash Memory Summit 2016Santa Clara, CA12
Cassandra on NVMf storage Widely used open-source NoSQLWe know that NVMe drives deliverimproved performance & latency However, NVMe drives areunderutilized (IOPS & Bandwidth) Can we use NVMf to deliver moreefficient remote storage ?Performance Analysis of NVMe SSDs andtheir Implication on Real World DatabasesFlash Memory Summit 2016Santa Clara, CAhttps://www.cs.utah.edu/ manua/pubs/systor15.pdf13
System ConfigurationYCSB Clients10GbeCassandra Docker ApplicationServerDell 730 Dual40GbeNVMf Target Storage ServerDell 930 X86 Server 8x PM1725Flash Memory Summit 2016Santa Clara, CAYCSB Workload: WorkloadA, 50/50 read/update,zipfian distribution WorkloadD, 95/5 read/insert, uniformdistribution Record count: 100 Million records,100 GB in each database Client Thread count: 1614
Workload D R95%I5%Workload A R50/U50Cassandra Client performanceNVMe NVMf tracks DAS performance with minor differencesFlash Memory Summit 2016Santa Clara, CA15
NVMEoF Target Performance Low Utilization on TargetFlash Memory Summit 2016Santa Clara, CA16
Fast and Efficient Storage For Cassandra NVMf PM1725 enables highperformance, efficient disaggregatedstorage Drive higher-utilization of storagesystems and NVMe devices Call to action : Add reliability features to NVMf More performance improvementsNVMf enables high-performance, low latency remote storage for databasesFlash Memory Summit 2016Santa Clara, CA17
Conclusions RedisOnFlash PM1725 enables larger DBs with fewer servers Maintains consistent 1ms Latency Exceed 1000K ops/sec for 100B-1000B objects Cassandra PM1725 with NVMf target delivers a high performance and scalableNoSQL SolutionFlash Memory Summit 2016Santa Clara, CA18
Thank YouManu AwasthiZvika GuzQiumin XuFlash Memory Summit 2016Santa Clara, CA19
Build efficient remote storage for databases Flash Memory Summit 2016 Santa Clara, CA 2. NVMe Design Advantages Lower latency Direct connection to CPU's PCIelanes . Comparison to SATA Flash Memory Summit 2016 Santa Clara, CA 10 80/20 read-write ratio. DRAM or Flash? Performance and Capacity Flash Memory Summit 2016
