IBM Systems and Technology Group Thought Leadership White Paper

April 2013

Preferred reads with IBM FlashSystem

2

Preferred reads with IBM FlashSystem

IBM Systems and Technology Group

3

Introduction
Many enterprise data centers have adopted disaster recovery (DR) solutions based on array-based replication. These array-based replication schemes are central to the business, so much so that the business becomes locked-in to the products from the array manufacturer. Businesses with array-based replication are reluctant to invest in alternative storage technology because introducing new products would lead to a fork-lift upgrade at multiple data centers. Preferred reads are a method to integrate solid state disks into these data center that takes advantage of existing array-based replication mechanisms, delivers extreme read performance benefits and maintains availability. Preferred reads are implemented at the file system or on a virtualization switch and provide mirrored writes and preferred reads. Writes are mirrored between the RAID system (which is providing array-based replication) and a solid state disk. Preferred reads come from a specified storage device (such as a solid state disk) or from the storage device with the shortest read queue (the solid state disk). Preferred-plex is a feature offered with Veritas. Preferred reads are also offered with IBM® and HP operating systems.

Brief discussion of array-based replication
Two mega trends have driven data center architectures over the last 10 years: increasing adoption of disaster recovery and increasing data center consolidation. The intersection of these trends has been at the storage array. Array manufacturers, such as EMC® and IBM, were first to the forefront with array offerings that provided massive capacities of storage and the ability to remotely replicate that storage to DR sites. These same trends are increasingly causing performance degradation. The array controller has to add DR awareness to its already complex caching and storage rules. Hard disk drives are getting denser to handle the surge in digital content and arrays are managing petabytes of storage capacity. Simultaneously, software is becoming more powerful and easier to use, resulting in surging user activity. The combination of these trends is over-loaded centralized storage systems with inadequate performance characteristics. Over the last few years, the number of options for DR has grown dramatically, but many companies are locked into array-based replication for the foreseeable future. It is enormously complicated and costly to stop using array-based replication or use a new vendor’s array-based replication.

Companies seeking to improve application performance in these environments are between a rock and a hard place. What is needed is a method to accelerate application performance without breaking the disaster recovery process.

Solid state disks have the following characteristics:

About flash storage
Flash storage is a proven technology. IBM® leads the industry with the most comprehensive flash portfolio with its acquisition of Texas Memory Systems, an IBM Company that has designed and manufactured solid state disks for over 30 years. Solid state disks are non-volatile storage devices that use RAM as the primary storage media. Solid state disks store and access data directly on DDR RAM chips, which results in storage speeds far greater than conventional, magnetic storage devices. Non-volatility is achieved through the integration of redundant internal battery systems and disk backup systems.

Lowest possible access times Flash-based solid state disks have access times below 25 microseconds, the best access times of any storage platform. As a comparison, a hard disk drive at 5 milliseconds has 333 times higher access times. High bandwidth The enterprise solid state disk market includes products, such as IBM FlashSystem 820, which can support 5.0 GB per second of random data throughput. High I/Os per second (IOPS) Flash storage extraordinarily high random I/O performance because of their low access times (latency). High availability Flash storage is inherently more reliable than hard disk drive based systems because their data path does not require moving parts. IBM FlashSystem also incorporates redundant components and patented Variable Stripe RAID™ technology to deliver no single point of failure within FlashSytem 720 and FlashSystem 820.

Figure 1: IBM FlashSystem™ 820

4

Preferred reads with IBM FlashSystem

IBM Systems and Technology Group

5

Non-volatile IBM FlashSystem arrays is a non-volatile solution. These systems have internal batteries that maintain system power until any volatile data can be written to flash.

Flash storage is an excellent solution for I/O bottlenecks, particularly those bottlenecks caused by the high access times of traditional disk based storage systems.

lowest. The default logical volume manager (LVM) of AIX and HP-UX function this way. This type of mirror dynamically chooses which storage system to read from based on the queue of outstanding IO requests to each half of the mirror while still mirroring writes. When two storage systems with different performance characteristics are mixed in this type of mirror, the faster storage system will end up receiving the majority of the IO due to its consistently lower IO queue.

Cost keeps a lot of companies from investing in solid state storage. The cost of mirroring solid state disk to another solid state disk scares away even more buyers. Preferred reads allow a company to avoid mirroring solid state storage (for reliability purposes) by mirroring to the existing RAID array thus minimizing the required investment. Fear keeps a lot of companies from investing in flash storage systems. One of the biggest fears is how flash can affect the availability of an application and the impact on DR operation. Implementing flash storage with preferred reads allays these fears. Here are some scenarios and how the data center is impacted: 1. Flash array is mirrored to an existing disk-based RAID array and in an extremely unlikely event, the flash fails catastrophically. In this scenario, the operating system and file system takes over and all writes and reads are from the disk-based RAID. Performance is no worse than it was prior to flash installation. 2. Disaster strikes and the main data center is out of commission. All data written to the flash has been written to the RAID array which has been replicating its data to the DR site. The DR site picks up operation as it was intended.

Performance availability
For some customers, combining a flash based FlashSystem 820 and existing disk-based RAID is a perfect blend of cost, performance and reliability. Others require additional levels of performance. Two methods are suggested for companies who cannot afford the drop in performance that could occur should a solid state disk become unavailable: 1. Create a triple mirror. Three writes are created. One write for each of two solid state disk arrays and one write for the RAID array. Preferred reads cascade from the first flash array to the second flash array and finally to the disk-based RAID. 2. Add flash to the DR site. Companies who fail-over to a DR site during routine maintenance or who require high performance during DR operations should implement flash into the DR site

Preferred reads in AIX
In AIX, preferred read is setup by traditional mirroring of logical volume manager (LVM) with a scheduling policy set to parallel. This will selectively read from the faster FlashSystem while the writes are synchronously written to both FlashSystem and traditional array. This function was originally created to provide a basic DR function by providing synchronous writes between two physically separated devices but allow preferred reads from the closest device in order to maintain high read performance.

Using preferred reads with flash storage
When preferred reads are used with a flash array, the outcome is application performance improvement. With preferred reads, all reads are from the flash array while all of the writes are written synchronously to both the flash array and the diskbased RAID. The DR features of the RAID will manage as before and replicate all writes to the DR site. This is an ideal set up for an application that is high read and low write. This approach will not improve performance for 100 percent write applications since writes only go as fast as the slowest device in the mirror. A tertiary benefit of this approach is that the disk-based RAID array can focus its performance on writes or other mission critical applications while the heavy read load is off-loaded to the solid state disk array.

Preferred reads based on queue depth
Some operating systems do not allow you to designate a specific drive for preferred reads, instead they conduct reads from the fastest storage device. The fastest storage device is determined by calculating which read queue to a device is

Summary
Array-based replication is a center point of many enterprise data centers. Companies with array-based replication find it difficult to change storage vendors or even add other storage devices, such as flash, to their environment. Software and operating systems that implement preferred reads are an important way to incorporate flash into the enterprise. Performance is increased, the DR site is still fed by the replicating RAID array and data center reliability is not compromised.

© Copyright IBM Corporation 2013 IBM Corporation Systems and Technology Group Route 100 Somers, NY 10589 Produced in the United States of America April 2013 IBM, the IBM logo, ibm.com, FlashSystem and Texas Memory Systems are trademarks of International Business Machines Corp., registered in many jurisdictions worldwide. Other product and service names might be trademarks of IBM or other companies. A current list of IBM trademarks is available on the web at “Copyright and trademark information” at ibm.com/legal/copytrade.shtml This document is current as of the initial date of publication and may be changed by IBM at any time. Not all offerings are available in every country in which IBM operates. The performance data discussed herein is presented as derived under specific operating conditions. Actual results may vary. THE INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS” WITHOUT ANY WARRANTY, EXPRESS OR IMPLIED, INCLUDING WITHOUT ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND ANY WARRANTY OR CONDITION OF NONINFRINGEMENT. IBM products are warranted according to the terms and conditions of the agreements under which they are provided. Actual available storage capacity may be reported for both uncompressed and compressed data and will vary and may be less than stated. Please Recycle

For more information
To learn more about IBM FlashSystem please contact your IBM sales representative or IBM Business Partner, or visit the following website: ibm.com/storage/flash

TSW03228-USEN-00

Sign up to vote on this title
UsefulNot useful