AWS Storage Tiers for Enterprise Workloads - Best Practices (STG301) | AWS re:Invent 2013


Published on

Enterprise environments utilize many different kinds of storage technologies from different vendors to fulfill various needs in their IT landscape. These are often very expensive and procurement cycles quite lengthy. They also need specialized expertise in each vendor's storage technologies to configure them and integrate them into the ecosystem, again resulting in prolonged project cycles and added cost. AWS provides end-to-end storage solutions that will fulfill all these needs of Enterprise Environments that are easily manageable, extremely cost effective, fully integrated and totally on demand. These storage technologies include Elastic Block Store (EBS) for instance attached block storage, Amazon Simple Storage Service (Amazon S3) for object (file) storage and Amazon Glacier for archival. An enterprise database environment is an excellent example of a system that could use all these storage technologies to implement an end-to-end solution using striped PIOPS volumes for data files, Standard EBS volumes for log files, S3 for database backup using Oracle Secure Backup and Glacier for long-time archival from S3 based on time lapse rules. In this session, we will explore the best practices for utilizing AWS storage tiers for enterprise workloads.

Published in: Technology, Business
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

AWS Storage Tiers for Enterprise Workloads - Best Practices (STG301) | AWS re:Invent 2013

  1. 1. STG301 - Amazon Web Services Storage Tiers Headline goes here Enterprise Workloads - Best Practices Sub headline goes here Tom Laszewski Solution Architect, AWS November 15, 2013 Thiru Sadagopan Cloud Services, Apps Associates © Copyright 2013. Apps Associates LLC. Chris Gattoni IT Director, RISO 1
  2. 2. Agenda • • • • • AWS storage options Enterprise storage challenges AWS storage solutions – Best practices, performance A storage configuration for Oracle Database Case study - RISO © Copyright 2013. Apps Associates LLC. 2
  3. 3. AWS Storage Options © Copyright 2013. Apps Associates LLC. 3
  4. 4. Storage Services on AWS • AWS provides multiple storage options for enterprise workloads Traditional workloads Storage migration techniques Application-specific storage options • • • • • • • • • • • • • • Ephemeral storage Amazon Elastic Block Store Amazon Simple Storage Service (S3) Amazon Glacier Amazon Direct Connect Amazon Export/Import Amazon Storage Gateway 3rd party storage gateways Tsunami, BitTorrent, 3rd party solutions Amazon ElastiCache Amazon DynamoDB Amazon RDS Amazon Redshift Amazon CloudFront © Copyright 2013. Apps Associates LLC. 4
  5. 5. Performance Is Critical • • • • What storage option has a transfer rate of 10s of MB/sec? What storage option has up to 32 MB/sec bursting to 40 MB/sec? 12 PIOPS volumes at 500 IOPS will saturate this type of instance This storage device has single-threaded performance that is limited to ~17MB/sec • What is the storage option that produces ~120,000 random read IOPS (4 KB blocks)? • A NoSQL database executing 4K random reads has performance of 3 MB/sec on this type of storage and ~200 to 260 MB/sec on this type of storage • Transfer ~3 TB a day using this storage option • What storage option produces ~100-145 MB/s read and write? © Copyright 2013. Apps Associates LLC. 5
  6. 6. Enterprise Storage Challenges © Copyright 2013. Apps Associates LLC. 6
  7. 7. Enterprise Storage Challenges Performance • Requires consistent performance, each application has unique requirements • Addressed by storage layers like SSD, SAN, NAS etc.. NAS, etc. Durability and availability Disk Backups, Tape Backups • Addressed by using RAID, disk backups, tape backups Cost Elasticity and scalability • Requires high CAPEX • Often addressed by over provisioning overprovisioning © Copyright 2013. Apps Associates LLC. 7
  8. 8. Enterprise Storage on AWS Performance Durability and availability Cost Elasticity and scalability • • • Requires consistent performance, each application has unique requirements EBS optimized instances Provisioned IOPS, SSD, Addressed by storage layers like SSD, SAN, NAS etc.. • Built-in replication across AZs; snapshots can be replicated • Addressed by using RAID, Disk Backups, Tape Backups across regions • Requires highyou use, buy vs. build Pay for what CAPEX • Often addressed by over provisioning need Use as little or as much as you want or © Copyright 2013. Apps Associates LLC. 8
  9. 9. Enterprise Use Cases – Storage Requirements Databases Application software File Server Archival storage (longterm backups) Latency (performance) Low Moderate Moderate Moderate Durability and Availability High High High – moderate High durability – Low Cost ($/GB) High Moderate Low Very low Elasticity and scalability High – predictable Low – stable High – spiky Medium – predictable © Copyright 2013. Apps Associates LLC. 9
  10. 10. AWS Storage Solutions © Copyright 2013. Apps Associates LLC. 10
  11. 11. Ephemeral Store – Best Practices • Use with caution • Not all instance types have instance store storage – Second-generation high-memory instances, micro instances • Use rsync to copy data from instance store volumes to EBS volumes as applicable to ensure protection • Initialize the volume before using for high-performance applications – dd if=/dev/zero of=/dev/sdb bs=1M • Add to AMIs explicitly if required, when creating images • Enable “termination protection” at instance level as an additional measure © Copyright 2013. Apps Associates LLC. 11
  12. 12. Ephemeral Store – Anti-patterns • Persistent storage – use Amazon ELB/Amazon S3 instead • Database storage – use Amazon ELB/PIOPS – Amazon ELB volumes – With proper bootstrapping, ephemeral store can be used for temporary table spaces – cr2.8xlarge/ hi1.4xlarge provide SSD backed instance-store volumes, which may be used for temporary database storage using specific database features • Shared storage – use Amazon ELB/Amazon S3 instead • Snapshots— use Amazon ELB instead – Optionally rsync the contents of instance store volumes to Amazon ELB volumes, depending on the use case © Copyright 2013. Apps Associates LLC. 12
  13. 13. Amazon EBS – Best Practices • Use striping for multiple volumes to increase IOPS – RAID 0 and RAID 1+0 RAID 0 ( Stripe ) RAID 1+0 ( Stripe and mirror ) Usage drivers I/O performance is more important than fault tolerance Fault tolerance is more important than I/O performance Advantages I/O is distributed across the volumes in a stripe, resulting in better performance Data is replicated and striped across volumes for better durability Disadvantages Overall performance of the stripe is limited to the worst-performing volume in the set Doubles the amount of I/O required from the instance to Amazon ELB Recoverability Amazon ELB snapshots are complex to manage, must depend on third-party backup methods Fault tolerant due to replication; however, Amazon ELB snapshots may be complex to manage and perform complete disk group recovery Cost Very little to no overhead; utilization ratio can be high Doubles storage cost. May also have impact on network performance at instance level during highvolume data transfers © Copyright 2013. Apps Associates LLC. 13
  14. 14. Amazon EBS – Best Practices • Use Amazon EBS-optimized instance types – Uses an optimized configuration stack that provides additional dedicated capacity for Amazon EBS I/O – Provides the best performance for Amazon EBS volumes by minimizing contention between Amazon EBS I/O and other traffic – Delivers dedicated throughput to Amazon EBS, with options between 500 Mbps and 1000 Mbps – Costs an additional low, hourly fee for the dedicated capacity – Cannot be enabled for a running instance; requires start and stop of instance – Not available for all instance types – Not required for cluster instance types © Copyright 2013. Apps Associates LLC. 14
  15. 15. Amazon EBS – Best Practices • Availability of Amazon EBS–optimized instance types Network performance EBS optimization Instance types Comments Very low / Low Not available m1.small, t1.micro Not recommended for I/O-intensive workloads; application scaleout is recommended. No m1.medium, c1.medium, m2.xlarge, Not recommended for I/O intensive workloads. Yes m1.large, m3.xlarge, m2.2xlarge Can provide up to 500 Mbps of dedicated Amazon EBS throughput. Can be used for moderate to high I/O workloads. High Yes m1.xlarge, m3.2xlarge, c1.xlarge, m2.4xlarge Can provide up to 1000 Mbps of dedicated Amazon EBS throughput. Can be used for high I/O workloads. 10 gigabit Not available cc2.8xlarge, cr1.8xlarge , hi1.4xlarge, hs1.8xlarge , cg1.4xlarge Amazon EBS optimization is not required, as these instances provide better network throughput and can deliver better I/O performance. Moderate © Copyright 2013. Apps Associates LLC. 15
  16. 16. Provisioned IOPS – Best Practices • Provisioned IOPS volumes are designed to deliver within 10% of the provisioned IOPS performance 99.9% of the time – When attached to Amazon EBS–optimized instances – Enough I/O requests to maintain its optimal average queue length – The read and write operations have a block size of 16 KB or less • Prepare new PIOPS volumes by touching all blocks to ensure consistent performance • Avoid taking Amazon EBS snapshots (for both PIOPS and Standard EBS volumes) during peak load • Avoid throttling overall network pipe/Amazon EBS optimization by attaching too many PIOPS volumes © Copyright 2013. Apps Associates LLC. 16
  17. 17. Amazon EBS – Performance • Queue length – Maintaining the optimal average queue length is crucial for delivering provisioned throughput – If the workload maintains an average queue length that is higher than your optimal value, then your per-request I/O latency will increase, and in this case you should provision more IOPS for your volume – Maintain average queue length of 1 for every 200 Provisioned IOPS and tune that value according to your application requirements – VolumeQueueLength Amazon CloudWatch metric can be used for each volume to identify potential performance issues © Copyright 2013. Apps Associates LLC. 17
  18. 18. Amazon EBS - Performance • Amazon CloudWatch metrics to monitor for Amazon EBS volumes – – – – – – VolumeQueueLength VolumeIdleTime VolumeReadOps VolumeWriteOps VolumeThroughputPercentage (for PIOPS volumes) VolumeConsumedReadWriteOps (for PIOPS volumes) • VolumeWriteOps/VolumeReadOps needs to be monitored for standard Amazon EBS volumes to find potential need to convert them to PIOPS © Copyright 2013. Apps Associates LLC. 18
  19. 19. Amazon EBS – Anti-patterns • EBS storage is not recommended for these use cases – Temporary storage • Must never use this for SWAP disks – Highly durable storage • Amazon EBS is designed to provide 99.5% – 99.9% annual durability. For better durability, use Amazon S3 – Static data or web content • Use Amazon S3 or Amazon CloudFront instead – Key-Value pair information • Use an AWS database service – Data that required very low access latency (like session data) • Use in-memory options such as Amazon ElastiCache or Amazon DynamoDB © Copyright 2013. Apps Associates LLC. 19
  20. 20. Amazon S3 – Best Practices • Amazon S3 performance considerations – – – – Amazon S3 can support 100 requests per second without much tuning effort If higher performance is required, optimization is required For GET-intensive workloads, use Amazon CloudFront For a mix of request types, optimize the object keys for better performance • For enterprise workloads – To optimize backup processes • Use multiple backup threads – If Amazon S3 backup is performed from within a VPC, make sure that Amazon S3 can be accessed directly from a private subnet • Usage of NAT instance may become a bottleneck © Copyright 2013. Apps Associates LLC. 20
  21. 21. Amazon S3 – Anti-patterns • Use cases where Amazon S3 is not recommended – File system • Even though it can emulate file system formats using keys, Amazon S3 is not recommended as an alternative to a file system. Using tools that can mount an Amazon S3 bucket as a file system may have severe performance/cost impacts – Structured data with query • Use Amazon Relational Database Service instead – Rapidly changing data • Use Amazon EBS volumes instead – Data that is rarely accessed • Use Amazon Glacier instead © Copyright 2013. Apps Associates LLC. 21
  22. 22. AWS Storage Options Summary Storage option General Database Applications Speed Instance Store Swap Temporary files Reports cache, web server cache Very low latency Very low; volatile Amazon EBS – PIOPS Data files, redo logs Durability Low latency Moderate; needs to be backed up regularly Amazon EBS Boot volume Binaries, archive logs Binaries Moderate latency Moderate; needs to be backed up regularly Amazon S3 Backups Backups Backups Longer latency Very high durability Amazon Glacier Long-term backups Long-term backups Long-term backups Restore times of Very high durability 3–5 hr © Copyright 2013. Apps Associates LLC. 22
  23. 23. A Storage Configuration for Oracle Database © Copyright 2013. Apps Associates LLC. 23
  24. 24. Database I/O Patterns • Cache data in memory from disk to achieve fast query response • For OLTP/DSS workloads, memory should be big enough to hold as much as possible SGA Buffer Cache 100% cache hit ratio is ideal for OLTP workloads © Copyright 2013. Apps Associates LLC. 24
  25. 25. Database I/O Patterns • CPU time is consumed mainly by I/O wait Disk I/O Time CPU Time Select Update Cache miss Select Update Cache hit © Copyright 2013. Apps Associates LLC. 25
  26. 26. Database Cache Has Limits 1. Data size is increasing... 1. Huge amounts of data 2. Not possible to cache all data SGA 2. User volume is increasing... Buffer Cache 3. Many I/O operations Source: oracle corporation © Copyright 2013. Apps Associates LLC. 26
  27. 27. Impact of I/O on CPU Response Times I/O bottleneck Low throughput and slow response Low CPU usage 700 16.00 14.00 600 12.00 TPS 500 RES 10.00 400 8.00 300 6.00 200 4.00 100 2.00 0 Response Time(Normal100Thread=>1.00) TPS(Normal-100Thread=>100) 800 0.00 Thread 100 200 300 400 500 600 700 800 Source: oracle corporation © Copyright 2013. Apps Associates LLC. 27
  28. 28. Storage Best Practices for Oracle • Oracle Automatic Storage Management feature is recommended option to optimize storage performance on AWS • SAME (Stripe and Mirror Everything) feature of Oracle ASM will help distributing entire data across multiple Amazon EBS volumes • ASM along with PIOPS Amazon EBS volumes can provide robust storage performance on AWS • Oracle Secure Backup for Cloud Module will enable backups directly from Oracle Database/ASM to Amazon S3 using RMAN • RMAN channels can be optimized for optimal system throughput and storage performance © Copyright 2013. Apps Associates LLC. 28
  29. 29. Typical Storage Configuration for Oracle Storage Type Comments Binaries Amazon EBS volume Standard volume; Amazon EBS snapshots enabled Data files PIOPS – Amazon EBS volumes Striped across multiple volumes using ASM or any other technology PIOPS of 1000 or above Redo log files PIOPS – Amazon EBS volumes Use separate Amazon EBS Volumes for each group; preferable to use ASM disk groups Archive log files Amazon EBS volumes Standard volumes for normal database workloads; PIOPS for highly transactional environments Backup files Amazon EBS volumes / Amazon S3 Standard Amazon EBS volumes for local backups; use OSB/other technology to push to Amazon S3 © Copyright 2013. Apps Associates LLC. 29
  30. 30. Storage Options for Oracle • Sample disk layout for Oracle database using ASM DEV PROD +PROD_DATA Striping +PROD_FRA Striping 4 EBS Volumes 4 EBS Volumes 1000 PIOPS 4000 PIOPS ASM instance with 16,000+ IOPS TEST +DEV_DATA S3 Bucket Lifecycle policies Striping 4 EBS Volumes 500 PIOPS Amazon Glacier © Copyright 2013. Apps Associates LLC. 30
  31. 31. Database Backup Best practices PROD +PROD_DATA +PROD_FRA Striping 4 EBS Volumes 1000 PIOPS /backupfs OSB cloud module using RMAN EBS snapshot S3 Bucket Striping 4 EBS Volumes 4000 PIOPS RMAN local backup © Copyright 2013. Apps Associates LLC. 31
  32. 32. SSD Storage with Oracle • AWS provides 2 instances types with SSD storage – cr1.8xlarge – 240 GB of SSD – hi1.4xlarge – 2048 GB of SSD • This SSD storage is ‘ephemeral’ and CANNOT be used for data files • However, this SSD can be leveraged by using “Oracle Flash Cache” – Available at no additional cost, Database Smart Flash Cache on Oracle Solaris and Oracle Enterprise Linux • Oracle Flash Cache simply extends the buffer cache to leverage SSD • Recommended use case for environments with huge working datasets – ETL processes for data warehouses – Reporting work load © Copyright 2013. Apps Associates LLC. 32
  33. 33. Case Study - RISO Chris Gattoni, IT Director © Copyright 2013. Apps Associates LLC. 33
  34. 34. About RISO, Inc. Who RISO Is Mission RISO means “ideal.” RISO strives to create fundamentally unique new technologies in paper-based communication. Every product we manufacture offers the ideal solution to our customers. At the same time, RISO carries on a tradition of caring for the world at large through our environmental and community support programs. Just as we are committed to delivering the ideal products and service, we strive to be an ideal corporate citizen in every country in which we conduct business. Company Overview RISO, Inc. is a world leader in digital printing solutions, with award winning printing technologies that are at the forefront of the printing equipment industry. RISO printers help improve productivity, increase versatility, and reduce costs.
  35. 35. About RISO, Inc. What RISO Does RISO sells high speed inkjet printers and digital duplicators. They're low-cost, durable, & reliable production printers. RISO strives to create fundamentally unique new technologies in paper-based communication through an approach that emphasizes productivity, cost containment, and versatility. Every product we manufacture delivers these benefits in a balance that offers the ideal solution to our customers. Speed up your color communications!
  36. 36. Key Business Drivers • Key challenges – – – – End-of-life hardware CAPEX vs. OPEX Upcoming datacenter migration IT not spending enough time on core functions • AWS solution – – – – Migrate from existing hardware to Amazon EC2 platform Lower TCO No datacenter to maintain , resulting in more flexibility for infrastructure team Scalable and elastic platform © Copyright 2013. Apps Associates LLC. 36
  37. 37. On-Premises Environment • Infrastructure spread across multiple locations – Connectivity required from multiple locations – Multiple servers, SAN Storage, network equipment – Printers, tape libraries, file servers • Applications include – – – – – Oracle ERP Oracle Business Intelligence LACROSSE applications on Microsoft SQL Server Domain controllers, PING identity, monitoring servers Ticketing Systems © Copyright 2013. Apps Associates LLC. 37
  38. 38. Target AWS Environment • Virtual Private Cloud with multiple VPN tunnels for redundancy – Separate VPN tunnels to each remote location – Access restricted only via VPN tunnel for better security • EC2 servers with Amazon EBS storage – Amazon EBS – PIOPS for database storage • Tape library replaced with Amazon S3 • DR setup for core infrastructure components like domain controller • DR build for enterprise applications in separate region © Copyright 2013. Apps Associates LLC. 38
  39. 39. Storage Options on AWS • Leveraged all storage options to ensure optimal performance in AWS – – – – Ephemeral – Swap space/Temp storage Amazon EBS volumes – OS boot volumes / Software binaries / Application storage Amazon EBS - PIOPS Volumes – Oracle data files, redo logs on striped volumes Amazon S3 (RRS) – For storing non-production backups using third-party backup appliance – Amazon S3 – For storing production backups – Amazon Glacier – Long-term archival of backups – AWS Export/Import – To transfer 2+ TB data to Amazon Glacier using USB disks © Copyright 2013. Apps Associates LLC. 39
  40. 40. Key Challenges • 32-bit vs. 64-bit instance types – – – – – AWS provides limited options to run 32-bit applications Legacy applications were not certified to run on 64-bit instances Upgrade to latest versions was not feasible 32-bit instance types were unable to provide enough throughput Network performance/Amazon EBS performance was insufficient to meet user load • Solution: Amazon EBS PIOPS with scale out – Implemented multiple Amazon EBS PIOPS volumes with Amazon EBS optimization for better I/O – Scaled out application middle tier to use multiple servers for better throughput – Used custom-built AMIs to run legacy applications to avoid AWS dependency © Copyright 2013. Apps Associates LLC. 40
  41. 41. Key Takeaways • • • • AWS provides multiple storage options for enterprise workloads No single option fits all workloads Proper planning and validation of I/O performance is the key for success Phased approach for AWS migration – – – – Start with a POC Migrate test/dev environments Build DR in cloud Migrate production workloads • Continuous monitoring of key performance metrics is crucial for success © Copyright 2013. Apps Associates LLC. 41
  42. 42. Resources • AWS Architecture Center • Storage Options in the AWS Cloud • AWS: Overview of Security Process • Roadmap to Moving Your Applications to the Cloud http:// • AWS Test Drive Labs – Try your application in the Cloud © Copyright 2013. Apps Associates LLC. 42
  43. 43. Connect With Us • Web: • Email: • YouTube: • LinkedIn: • Twitter: @AppsAssociates • Facebook: © Copyright 2013. Apps Associates LLC. 43
  44. 44. Thank You!