Flash for the Real World – Separate Hype from Reality

Editor's Notes

• #6: EMC will announce this calendar quarter their new VMAX II; rest assure we are ready to compete with the VSP.Here are some of the key enhancements which we will deliver this calendar year to give you the upper hand:More flash options, starting with Multi layer (MLC) SSD this month and additional MLC flash based product later on this year A new VSP microcode with a huge performance and scalability boost, we called it flash acceleration. This enhancement will be deliver to our existing customer
• #8: Current SSD vendors are focused on the volume market (laptop and PC) which is tied to the 2.5 inch and 3.5 inch form factor….. Their focus is on lowest cost per unitOur focus is for a different use: for enterprise workloads and data center applications. Our focus is on true sustained performance (Random reads/writes, Sequential), lowest bit cost, capacity and extended endurance.
• #9: Explicit
• #13: Explicit
• #14: Explicit
• #15: Explicit.Superior data integrity is designed in the flash controller, each FMD ECC can correct 48 bits per 1.4KB vs MLC NAND specs of 24 bits per 1KB.
• #17: Explicit
• #18: Hitachi unique flash module packaging offers 4x the capacity of 400GB SSDWhen used in the storage system, these flash modules double the density per rack unit over solid state drives
• #19: But Hitachi flash innovation extend well past our Flash controller and up into our HUS VM and VSP systems. While the SPC performance numbers that Silverton Consoluting mentioned earlier in the presentation already demonstrated the high performance of our HUS systems, the architectures of the VSP and HUS VM allow for even greater performance scalability and customization.For years, storage systems were built with the idea that the performance was held back by spinning disk drives, and system software was optimized to make the most out of that situation. With flash, vendors are rushing to re-architect key parts of their onboard system code to more quickly move data around the inside of storage systems to keep up with the high performance of Flash. Hitachi, however, isn’t scambling to deliver this customization. We already have.VSP optimization was delivered in 2012 and HUS VM will be delivered in June 2013. The system code for these arrays see significant flash-focused re-engineering, with over 30 fundamental changes to the code to enable higher thread-count processing and faster movement of data. Response times improve significantly and provisioning and functions become twice as useful.Of course, in staying with Hitachi’s focus on high-availability, these changes are non-distruptively installed and become transparent to other installed storage replication or management tools.
• #21: To match the growth of your company’s application data, we give you several ways to scaleDynamic provisioning creates large, dynamic storage pools of dataDynamic tiering automates data placement of both internal and externally attached dataTo meet the workload demands of your virtual servers, both block and file data may be provisionedAnd you can scale out to manage, provision and reclaim space in external virtualized storageVery high density storage can be configured with dense trays and SSD options
• #23: Our method of data migration is the lowest risk method availableVirtualizing storage reduces outage window to applications to speed up technology refreshIn fact, data migration can be performed with 80% less effort then host or replication based methodsHost and replication methods can take up to 3 hours: estimating 150 minutes makes the virtualization method 80% less (150-30/150=.8)Because the storage controller is moving data in the background, there is no impact to performanceOur Nondisruptive solution will eliminate the application downtime currently required without ever touching the host, supports clustered configurations and will be performed with more automation to lower the technology refresh effort and risk.Host based methods include migration using replication technology (not array-based replication) which involves replication of storage volumes. Volume management is used primarily to control disk resources by mapping the logical view of storage space with the actual physical disks. Host-based mirroring or replicationsolutions generally focus on file-by-file data movement to create asecondary data copy for disasterrecovery purposes. Using server virtualization such as VMware VMotion to assist with servermigrations that can assist withmigrating the data as well. It is usedfor a storage area network (SAN)environment that is supported byVMware and when migrating aVMware virtualized environment. Host-basedmigrations that leverage NAStechnology for file-based migrationseither through NFS or throughCIFS to migrate CIFS or NFS filesystems to or from an existingproduction file server NAS systemto another, files are moved basedon the network or file systemprotocols. File system-based data replication allows copyingor relocating of both file data andfile system metadata, dependingon the type of server operatingsystem, replication software or NASsolution. Replication can includenot only the data but also thepolicy, rules and schedules for thefile system environment.Source: Data Migration Best Practices and NondisruptiveMigration Service Capability for Enterprise Storage,February 2012http://ucmprodextranet.corp.hds.com/conx/groups/intranet/documents/contentasset/01_072393.pdf
• #24: Hitachi ShadowImage® Heterogeneous Replication software enables immediate use of data in decision support, software testing and development, and data protection operations independent of operating system, application or device. Hitachi Thin Image will be available in the first microcode update. Hitachi Thin Image replaces Copy On Write to provide logical snapshot data replication within Hitachi storage systems.Hitachi Replication Manager is application aware for Microsoft Exchange and SQL so you can restore application data knowing it is integrated with both applications for complete or tuned recovery.JetMirror: Object Based High Speed Replication is 26x faster than file replication with incremental changes and 2.8x faster than file replication on full file replication Ideal for DR and data migration w/small backup window,Achieved by eliminating need to lookup directories/filesUnlike block replication, does not send empty containers due to object-awareJetMirror: File Replication Option:Granular replication of file system directories, virtual volumes, or snapshotsPolicy-based management and schedulingleverages built-in snapshot and NDMP featuresPreserves user/group permissions and quota information,Asynchronous only over IP networksJetClone: Most Flexible and Space Efficient,16x more clones per file system. Reduces file system limitations inherent with competitors.Space efficient “file clones” for virtual desktop infrastructures (VDI) and large writeable snapshotsCreate More Clones-Don’t run out of space for cloning data due to File System limits.Clones only container files, not whole file systemA single File System holds thousands of container filesRapid CloningUses same source data as an existing volume Highly Space Efficient-Contains only pointers to shared blocks from the original container file-User adds only the unique data to a clone
• #25: Hitachi Universal Replicator will replicate any data from any storage system virtualized across any fibre channel distanceHitachi TrueCopy SynchronousRemote Replicationhost independent data replication automates the recovery process Based upon our virtualization technology, both HDS and non-HDS storage data may be replicatedAsynchronous, logged transmission ensures there are no interruptions in service from synchronization delaysThus allowing you to ensure automated recovery to meet your objectives without interruption
• #26: The100% data availability guarantee is available for Wister – no one else offers oneThe 100% data availability guarantee is an optional contract exhibit, not everyone automatically receives itThe terms are the same as VSP: HiTrack is required and so is an active service agreementWister is an entry enterprise storage system based on VSP architecture, it is not midrangeEntry enterprise storage requires enterprise availabilityThe file module is available in a clustered configuration for higher availabilityThe block module features shared processor and cache resourcesNew design integrating VSP functions into smaller package utilizing the HiStar Network architectureNot a scaled down VSP, an all new product specifically designed for this market segmentCache backup protection is onboard via flash device which is battery powered Hitachi HiTrack remote monitoring is available to oversee system functionsUpdates of components and microcode can be accomplished under operation
• #27: The building block of Hitachi Accelerated Flash storage is the flash module drive(FMD) to the left. At announcement the 1.6TB flash module is available, the 3.2TB FMD is scheduled for March 2013.The width of the FMD is roughly equivalent to a 3.5 inch LFF disk drive, except that it is much deeper. This new form factor optimize the depth of the rack which is currently not use in most storage chassis design using commodity form factor.This new design double the flash density currently possible with SFF SSD and quadruple the available capacity per device when compared to the 400GB MLC SSD currently shipped with our VSP.A new flash chassis is required to support these flash module; it fits in a standard 19 inch rack and will consume 8U of rack space with an extra 5U space needed for cabling. By default, the flash chassis will be installed 1st in the SAS chain, located just above the VSP controller chassis.
• #28: The new Hitachi Accelerated Flash storage option for HUS VM includes:The 1.6TB flash moduleUp to 12 flash modules fit in the 2U flash storage enclosureUp to 8 flash storage enclosures are supported per HUS VM system96 flash modules for 154TB raw maximum capacity in 16 rack units
• #29: Hitachi Accelerated Flash storage triples the economics over solid state based HUS VMLower cost per IO: 500,000 IOPS with flash = $1.03, SSD = $2.87 (list prices). 2.87-1.03=1.84, 1.84/2.87=64% improvement or 3xIt requires 80 SSD to achieve 500,000 IOPS while only 8 HAF modules can achieve 500,000 IOPS. HDD cannot achieve this performance.All prices are US list pricePhysical space savings are also substantial compared to SSD and HDDReference sources:http://www.thestoragealchemist.com/flash-is-really-about-economics/?sf12791614=1http://wikibon.org/wiki/v/Performance_Centric_Data_Storage_TCO_Study_Demonstrates_Cost_Advantages_of_Flash_Arrayshttp://wikibon.org/wiki/v/Case_Study:_The_Hunting_of_the_RARChttp://wikibon.org/wiki/v/Flash_and_Hyperscale_Changing_Database_and_System_Design_Forever
• #30: In this comparison the Violin Memory 16TB model 6616 (SLC) is compared to the HUS VM at 13TBAt list price they are fairly similar. The source for pricing is the Cisco TPC-C disclosure below.The source for the Violin performance data is the ESG lab validation below of two arrays. We have discounted the 4K IOPS by 40% for a single array to equal out the IOPS at an 8K block size.HUS VM was configured with 8 FMDs in 2 trays, 64GB cache, 2 FC I/O modules, 4 SAS I/O modulesWe’re giving Violin credit for 500,000 8K IOPS even though they claim 1 million 4K IOPS.Sources: Violin Memory TPC-C with Cisco http://c970058.r58.cf2.rackcdn.com/fdr/tpcc/Cisco-Oracle%20C240%20TPC-C%20FDR.pdfV-6000 Memory Array (V-6616-HA64-8xFC) $540,000 + $54,000 3 year bronze maintenance = $594,000 (10/26/12)ESG Lab Validation Violin Memory http://www.violin-memory.com/wp-content/uploads/ESG-Lab-Summary-Violin-6000-Series-Flash-Memory-Array-Aug-12-2pgr.pdf?d=11,352,467 4K IOPS with 2 Violin model 6616 arrays. 676,233 per array. Discounting the 4K workload by 40% = 405,740 8K IOPSViolin Memory published two performance results using Oracle to simulate 4K and 8K IO on their MLC flash product. The first in June 2012 and the other on August 2012While Violin Memory claimed that the 3205 model (MLC flash) is rated at 300K 4K IOPS, their true 8K performance is much less based on their first benchmark published by a factor of 40%. It is reasonable to assume a similar 40% decrease in the August 2012 results due to the same overhead associated with larger block size.http://loop.hds.com/thread/9494?tstart=0HDS HUS VM list price $513,721 for 8 FMD in 2 trays with 64GB cache.
• #31: Hybrid ArraysEMC VNX 7500 240,000 IOPSHP 3PAR StoreServ 10400 210,000 IOPSIBM XIV Gen 3 180,000 IOPSAll Flash ArraysViolin Memory 6616 SLC 1 million IOPSIBM FlashSystem 820 eMLC 525,000 IOPSPure Storage FA 400 400,000 IOPSEMC XtremIO X-brick 250,000 IOPS each
• #32: Starting from the bottom and working our way up:There are may types of flash storage offerings – every major storage vendor has one whether it is solid state drive based or unique all flash designsFewer of these flash offerings are multiprotocol or offer file integration as wellFew all flash vendors have sophisticated replication capabilities, it is the major storage vendors who can offer viable replication technologies for backup or disaster recoveryEven less of the storage offerings available have both thin provisioning and automated tiering. Although these features are standard now for the top storage systems, you will not find these capabilities in the all flash storage offered by the startupsExternal storage virtualization as we know is only available from HDS, EMC, IBM and NetApp with HDS being the leaderAnd finally, adding up all the capabilities, only HDS can offer an all flash unified array with external virtualization, remote and local replication, thin provisioning with automated tiering in a single storage system with one management platform
• #33: As we’ve discussed, HUS VM with flash delivers:Hitachi innovative Accelerated Flash Storage delivers higher performance for faster access … here HAF provides very low latencies.Better efficiency – more intelligent use of assets by placing frequently accessed data on HAF and data that is access less frequently on more cost effective HDDs.This unique storage system goes beyond flash. It is completely upgradable to a tiered storage system with hard drives, extends to third party storage with Hitachi external storage virtualization and may be upgraded to a unified platform to consolidate both block and file data.Hitachi delivers much more value than flash only storage systems. As an all flash system it accelerates application performance and may be upgraded for many more capabilities for consolidating file data, tiering data, disaster recovery, data mobility and analytics.Hitachi HUS VM All Flash = No Compromises Flash StorageHitachi Flash Acceleration software à Best Storage Optimization for FlashHitachi Accelerated Flash storage à Best Flash Form Factor and DurabilityUnified and Virtualized Capacity à More Flexibility Than Any Other All Flash system