Oracle Exadata: What's New And What's Coming - NoCOUG [PDF]

Léa Legrand | Download | HTML Embed

Jan 1, 1970
Views: 32
Page(s): 36
Size: 1.46 MB
Report

Transcript

2 Safe Harbor Statement The following is intended to outline our general product direcOon. It is intended for informaOon purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or funcOonality, and should not be relied upon in making purchasing decisions. The development, release, and Oming of any features or funcOonality described for Oracles products remains at the sole discreOon of Oracle. Copyright 2016, Oracle and/or its aliates. All rights reserved. | 2

3 Exadata Vision Drama.cally Be3er Pla5orm for All Database Workloads Ideal Database Hardware - Scale-out, database opOmized compute, networking, and storage for fastest performance and lowest costs Smart System SoWware specialized algorithms vastly improve all aspects of database processing: OLTP, Analy.cs, Consolida.on Full-Stack IntegraOon Database-to-disk opOmizaOon, automaOon, tesOng, patching, and support to reduce operaOonal costs Iden.cal On-Premises and Oracle Public Cloud Exadata Cloud Service Copyright 2016, Oracle and/or its aliates. All rights reserved. | 3

4 Proven at Thousands of CriOcal Deployments since 2008 Half OLTP - Half Analy.cs - Many Mixed 4 OF THE TOP 5 BANKS, TELCOS, RETAILERS RUN EXADATA Petabyte Warehouses Online Financial Trading Business Applica.ons SAP, Oracle, Siebel, PSFT, Massive DB Consolida.on Public SaaS Clouds Oracle Fusion Apps, Salesforce, SAS, Copyright 2016, Oracle and/or its aliates. All rights reserved. | 4

6 Ideal Hardware Architecture for Database Most Advanced - Highest Performance - Always Available - Starts Small, Scales Huge Ideal Server Architecture Scale-Out with Fastest CPUs Ideal Network Architecture Unied Ultra-fast InniBand Ideal Storage Architecture Scale-Out Servers + Database Ooad Ideal Flash Architecture Ultra-fast NVMe PCIe Flash Ideal Capacity Architecture Tier PCIe Flash & Huge 8TB SAS3 Disks PCI Flash Copyright 2016, Oracle and/or its aliates. All rights reserved. | 6

7 Exadata X6-2 Hardware Details (new in red) Database Server Scale-Out 2-Socket Database Servers 44 CPU cores Fastest Intel 22-core Broadwell: E5-2699 v4, 25% faster DDR4 DRAM frequency increased 13% Ultra-Fast Unied InniBand Internal Fabric Scale-Out Intelligent 2-Socket Storage Servers High Capacity (HC) Storage Intel 10-core Broadwell CPUs (25% faster) Ooad database processing 12.8 TB PCI Flash 96 TB disk 8TB Helium Disk Drives (2X Larger since Oct) 20 CPU cores 2X larger and 2X faster 3D V-NAND NVMe Flash cards Extreme Flash (EF) Storage High Capacity Rack - 1.3 PB Disk, 180 TB PCIe Flash 25.6 TB PCI Flash Extreme Flash Rack - 360 TB PCIe Flash 20 CPU cores Copyright 2016, Oracle and/or its aliates. All rights reserved. | 7

8 ElasOc ConguraOons Incrementally Scale Servers Achieve any Level of Performance with Minimum Hardware Database Server v w Incrementally Add Racks add DB or Extreme Flash Storage to Storage Con.nue Servers Scaling High Capacity Storage u Start Small Full Rack Mul.-Rack 2 Database Servers 3 Storage Servers Enable Database CPU cores as needed with Capacity on Demand Expand older Exadata machines with new X6-2 servers Copyright 2016, Oracle and/or its aliates. All rights reserved. | 8

9 Exadata X6 Delivers Breakthrough DB I/O Performance 301 GB/sec Analy.c Throughput 2.5X 5.6 Million 8K OLTP Read IOPS 5.2 Million 8K OLTP Write IOPS 250 us I/O latency at 2.4 Million IOPS (Scales higher as racks are added) Performance of 1 Exadata Rack with 10 DB servers and 12 Extreme Flash storage servers Copyright 2016, Oracle and/or its aliates. All rights reserved. | 9

10 Exadata X6 Pricing Unchanged to Greatly Improve Value 2X Huge benets over X5 for same price Performance 2X disk capacity, 2X ash capacity, 2X faster ash Value 25% faster X6-2 Broadwell CPUs Faster DRAM Minimum DB licenses unchanged (COD) 16 cores for Eighth Rack, 28 cores for Quarter Rack Capacity New X6-8 Exadata updated to X6 storage 2X Recovery Appliance (ZDLRA) updated to X6 Oracle SuperCluster updated to X6 storage Copyright 2016, Oracle and/or its aliates. All rights reserved. | 10

12 Smart System SoWware Highlights Smart Analy.cs Smart Storage Move queries to storage, not storage to Hybrid Columnar Compression reduces queries space usage by 10X AutomaOcally ooad and parallelize queries Database-aware Flash Caching gives across all storage servers speed of ash with 100X faster analyOcs capacity of disk PCI Flash Smart OLTP Smart Consolida.on Special InniBand protocol enables highest Workload priori.za.on from CPU to speed, lowest latency OLTP network to storage ensures QoS Ultra-fast transacOons using 4X more Databases in same hardware DB opOmized ash logging algorithms Fault-tolerant In-Memory DB by mirroring memory across servers Copyright 2016, Oracle and/or its aliates. All rights reserved. | 12

13 Smart System SoWware Introduced in 2015 Smart Analy.cs Smart Consolida.on 5X faster scans by converOng data to Zero overhead VMs Columnar format in Flash Cache Snapshots for test/dev 3X faster JSON/XML by Set ash cache min size per DB to ensure QoS ooading to storage servers VM VM InniBand parOOoning IPv6 for Ethernet Smart OLTP Smart Licensing 3X faster OLTP messaging using direct DB to Capacity-on-Demand reduces license cost by InniBand access disabling unneeded cores Instant detecOon of node failure Trusted Par..ons limit license scope of Sub-second capping of I/O latency by specialized opOons rerouOng I/Os to faster storage Copyright 2016, Oracle and/or its aliates. All rights reserved. | 13

14 Exadata Implements Much More Smart System SoWware Smart Analy.cs Smart OLTP Smart Availability Storage Index data skipping Ac.ve AWR end to end monitoring Priori.ze rebalance of cri.cal les Storage ooad for min/max Smart write-back ash cache Elimina.on of false drive failures opera.ons Cell-to-cell rebalance preserving Flash and disk life cycle mgmt alert Data mining storage ooad Flash Cache and Storage Indexes Avoid reading predic.ve failed Storage ooad for LOBs and Database scoped security disks CLOBs Full-stack security scanning Cell soUware transparent restart Auto ash caching for table scans Exachk full-stack valida.on I/O hang hardening Reverse ooad to DB servers Instant data le crea.on Prevent shutdown if mirror server Ooad index fast full scans Ac.ve bonding of InniBand is down Ooads scans on encrypted data, Disabling of unreliable network links Connement of temporarily poor with FIPS compliance performing drives Cri.cal DB messages always jump to head of queue for ultra-fast latency Data integrity valida.on (HARD) I/O issued by interac.ve users and Automa.c disk scrub and repair important workloads is priori.zed Copyright 2016, Oracle and/or its aliates. All rights reserved. | 14

15 Super Fast SoWware Updates 2.5X speed up in Storage Server SoWware Update Parallel rmware upgrades across components such as hard disks, ash, ILOM/BIOS, InniBand card Reduced reboots for SoWware updates Imaging speedup X5 High Capacity cell reimaged in less than 20 minutes Copyright 2016, Oracle and/or its aliates. All rights reserved. | 15

16 High Redundancy on Quarter and Eighth Racks Problem: On an eighth or a quarter rack the VoOng Disks were on a Normal Redundancy disk group and thus suscepOble to concurrent failures SoluOon: Create quorum disks on database servers in addiOon to those on storage servers Best PracOce: Use HIGH redundancy for DATA diskgroup and place voOng disk in HIGH redundancy diskgroup Oracle Exadata Deployment Assistant automaOcally creates quorum disks Quorum Disk Manager UOlity creates and manage quorum disks Minimum Grid Infrastructure SoWware version required: Oracle Database 12c Release 1 (12.1) release 12.1.0.2.160119 with these patches: 22722476 and 22682752 Copyright 2016, Oracle and/or its aliates. All rights reserved. | | Oracle CondenOal Highly Restricted 16

17 Storage Index PreservaOon across Rebalance In event of a disk failure data needs to be rebalanced Region Index out to disks on other cells A BB C D Previously, storage indexes created for the regions on Min B = 1 9 33 the failed disk were lost and recreated on the next scan Max B = 3 10 11 8 22 Storage index entries will be moved along with data to the new disk during cell to cell ooaded rebalance Min B = 8 11 99 10 88 Maintains applicaOon performance during rebalance Max B = 9 10 88 Minimum Grid Infrastructure soWware version: 6 Oracle Database 12c Release 1 (12.1) release 12.1.0.2.160119 with patch Min B = 5 7 22682752 Max B = 7 5 Copyright 2016, Oracle and/or its aliates. All rights reserved. | | Oracle CondenOal Highly Restricted 17

19 Exadata Full-Stack IntegraOon Reduces OperaOons Costs Unique Full-Stack Integra.on Unique Full-Stack Reliability All layers pre-congured, pre-tuned, All users run idenOcal full stack pre-debugged You get: DB, OS, drivers, rmware, Bank tested full-stack HA network, servers, storage Telco tested full-stack scaling Government tested full security Unique Full-Stack Support Unique Full-Stack Management One support team expert in and Full-stack management tool accountable for full stack Drill down from DB to Oracle performs free full-stack storage and up from updates and 24/7 monitoring storage to DB Copyright 2016, Oracle and/or its aliates. All rights reserved. | 19

20 Exadata Hardware GeneraOonal Advances* V1 V2 X2 X3 X4 X5 X6 Sept 2008 Sept 2009 Sept 2010 Sept 2012 Nov 2013 Dec 2014 Apr 2016 Xeon E5430 Xeon E5540 Xeon X5670 Xeon E5-2690 Xeon E5-2697v2 Xeon E5-2699 v3 Xeon E5-2699 v4 X6 Growth Q4, 2007 Intel GA Q1, 2009 Intel GA Q1, 2010 Intel GA Q1, 2012 Intel GA Q3, 2013 Intel GA Q3, 2014 Intel GA Q1, 2016 Intel GA Factor Storage (TB) 8 X Flash (TB) 32 X CPU (Cores) 5.5 X Memory (GB) 24 X Ethernet (Gb/s) 50 X * Assumes typical full rack congura6on of 8 database servers and 14 storage servers Copyright 2016, Oracle and/or its aliates. All rights reserved. | 20

21 Exadata Investment ProtecOon Expand Exis.ng Systems with X6-2 Servers - Run New SoUware on Older Hardware Database Server Upgrade older Exadata systems Extreme Flash Storage with X6-2 servers High-Capacity Storage Transfer soUware licenses at Run new soUware releases on earlier Exadata systems no cost EXTREME PERFORMANCE EXTREME PERFORMANCE DATABASE MACHINE DATABASE MACHINE DATABASE IN-MEMORY EXADATA X2 EXADATA X3 EXADATA X4 EXADATA X5 EXADATA X6 NON-EXADATA SYSTEMS Copyright 2016, Oracle and/or its aliates. All rights reserved. | | Oracle CondenOal Highly Restricted 21

22 The Exadata Community Eect Exadata Takes Standardiza.on to the Next Level Standard Technologies Oracle Public Cloud Oracle Development & Support Standard ConguraOons 1,000s of Customers and Partners Standard IntegraOon Standard Tuning Server Vendor A Standard Support Storage Vendor B Network Vendor C Database Vendor D The Scalability of Telecom, The Availability of Banking, OS Vendor E The Security of Government VM Vendor F The New Global Standard vs A Company Standard Copyright 2016, Oracle and/or its aliates. All rights reserved. | 22

23 Exadata Advantages Increase Every Year Exadata Cloud Service In-Memory Columnar in Flash Smart Fusion Block Transfer Drama.cally Be3er Pla5orm In-Memory Fault Tolerance for All Database Workloads Direct-to-wire Protocol JSON and XML ooad Instant failure detecOon Network Resource Management 3D V-NAND MulOtenant Aware Resource Mgmt Flash PrioriOzed File Recovery SoWware-in- IO PrioriOes Silicon Data Mining Ooad Tiered Disk/ Flash Ooad Decrypt on Scans Database Aware Flash Cache PCIe NVMe Flash Storage Indexes Columnar Compression Unied InniBand Smart Scan InniBand Scale-Out DB Processors in Storage Scale-Out Storage Scale-Out Servers Copyright 2016, Oracle and/or its aliates. All rights reserved. | 23

25 Shared Storage Has Many Advantages over Local Storage Servers Much bever space u.liza.on Much bever security, management, reliability SAN/LAN Enables DB consolida.on, DB high availability, RAC scale-out Shares storage performance Aggregate performance of shared storage can be dynamically used by any server that needs it Shared Storage Copyright 2016, Oracle and/or its aliates. All rights reserved. | 25

26 Flash Performance is Wasted by Shared Storage Arrays SAN Link = 40Gb New improvements in ash performance are causing 5 GB/sec 100X bo3lenecks across shared storage stack Latest PCIe Flash Speed of one ash card is now similar to fastest SAN or LAN link 5.4 GB/sec Throughput of a few ash cards is too fast to transfer to servers All-Flash Storage Array IO Path: many steps, each adds latency and creates bo3lenecks SSD Array Host Heads Switches CPU Ctrl HBA Flash Chips SAS/SATA SAN/LAN SAN/LAN PCIe Copyright 2016, Oracle and/or its aliates. All rights reserved. | 26

27 Exadata Achieves Memory Performance with Shared Flash Exadata Exadata X6 delivers 300GB/sec ash bandwidth to any DB Servers server Approaches 800GB/sec aggregate DRAM bandwidth of DB servers Query Must move compute to data to achieve full ash poten.al InniBand Ooad Requires owning full stack, cant be solved in storage alone Fundamentally, storage arrays can share ash capacity but not ash performance Even with next gen scale-out, PCIe networks, or NVMe over fabric Exadata Smart Storage Shared storage with memory-level bandwidth is a CPU paradigm change in the industry Flash PCIe NVMe Get near DRAM throughput, with the capacity of shared ash Chips Copyright 2016, Oracle and/or its aliates. All rights reserved. | 27

29 Users Demand Instant Fulllment of Any InformaOon Need Pa.ence is an ex.nct virtue Customers demand instant product informaOon and immediate acOon Prospects lose interest and move-on in seconds Employees demand instant results from internal applicaOons Finance Marke.ng Instant fulllment requires OLTP systems to deliver results in real-Ome, all-the-Ome Scale OLTP to any volume Human Resources Engineering Deliver instant results all-the-Ome Even when consolidaOng databases to reduce costs Sales Copyright 2016, Oracle and/or its aliates. All rights reserved. | 29

30 Exadata Uniquely Eliminates Bovlenecks to Scaling OLTP Exadata eliminates the tradi.onal OLTP bo3leneck: Random I/O PCI Flash Unique scale-out storage with ultra-fast PCIe ash, ultra-fast NVMe protocol, ultra-fast InniBand, and ultra-fast iDB protocol delivers: Over 5 Million DB reads or writes per rack; millisecond latency Unique Smart Flash Logging eliminates OLTP slowdowns and bovlenecks due to logging Exadata eliminates Inter-node coordina.on bo3lenecks for OLTP Unique Direct-to-Wire Protocol gives 3x faster inter-node OLTP messaging Unique Smart Fusion Block Transfer eliminates the need to write the log le when moving blocks between nodes Exadata delivers instant analy.c results directly on OLTP data Unique query ooad to storage, shared PCIe ash, InniBand, and In-Memory Database scaling Copyright 2016, Oracle and/or its aliates. All rights reserved. | 30

31 Exadata Uniquely Delivers OLTP in Real-Time, All-the-Time Exadata eliminates mixed-workload interference on OLTP opera.ons Unique OLTP message bypass of prior queued messages at host, switches, and storage Unique priori.za.on of OLTP I/Os from DB to disk bypassing I/Os from batch/reports/ analyOcs Unique storage ooad of RMAN backup Unique priori.za.on of opera.ons against important database vs less important Exadata eliminates OLTP stalls from failed or sick components Unique full-stack integra.on for fastest recovery of failed server, storage or switch/link Unique detec.on of server failures without a long Omeout avoids system hangs PCI Flash Unique sub-second redirec.on of I/Os around sick devices avoids database hangs Unique In-Memory Database Fault Tolerance Copyright 2016, Oracle and/or its aliates. All rights reserved. | 31

32 Exafusion Direct-to-Wire OLTP Protocol Worlds First Database to InniBand Protocol 3X Faster InniBand Exafusion InniBand has great throughput But overheard of calling through OS on every message limits small message rate Exafusion re-implements RAC Cache Fusion InniBand Database directly calls InniBand hardware Bypasses networking soWware stack, interrupts, scheduling 10 GigE 8K OLTP Block Transfers/sec Copyright 2016, Oracle and/or its aliates. All rights reserved. | 32

33 Exadata Smart Flash Logging Flash delivers great I/O throughput for OLTP Smart Logging = O Smart Logging = On But Flash has 100X spikes in response .me Outliers Response spikes are not a problem for most DB I/ Txn Response Time (ms) Os, but are big issue for OLTP log writes TransacOon commits must occur in order, so a slow Outliers log write stalls all transacOons Smart Flash Logging uses ash as a parallel write Outliers cache to disk controller cache or other ash No Outliers Write that completes rst wins (disk or ash) Shown to greatly improve OLTP throughput and Parallel response Omes in real workloads Log Writes (rst wins) Copyright 2016, Oracle and/or its aliates. All rights reserved. | 33

34 Smart Fusion Block Transfer OLTP workloads can have hot blocks that are frequently updated (e.g. right-growing index ) (Prior ) Inter-Instance Log le must be wriven before transferring a hot block Block Transfer Protocol between RAC instances so the block can be recovered 1. Issue log write Adds latency and reduces throughput On Exadata, Oracle does not wait for the log write ! Exadata ensures the log write completes before changes to 3. Transfer 2. Wait for log block on another instance commit, guaranteeing durability block write comple.on Wait for Log I/O during transfer of hot blocks is eliminated Up to 40% throughput and 33% response .me Exadata Avoids I/ improvement in some heavily contended OLTP workloads O Wait Copyright 2016, Oracle and/or its aliates. All rights reserved. | 34