Introduction Storage Area Networks dominate today s enterprise data centers. These specialized networks use fibre channel switches and Host Bus Adapters (HBAs) to connect to storage arrays. With software, these (oftenproprietary) SAN arrays can be networked for clustering, replication and sharing to achieve levels of reliability and scalability that are not available with Direct Attached Storage (DAS). One of the biggest advantages of SAN is the ability for multiple servers to share a large storage array. This sharing drives down cost and improves array utilization. Unlike DAS, where all of the capacity of the array is dedicated to a single server, SAN systems can be partitioned or even thin provisioned to support multiple server connections. Thin provisioning is a software technique deployed in the array that allows for capacity to be delivered to the application on-demand rather than over provisioning the array to meet future demands of the application. Now that PCIe Flash technology has become mainstream, thanks to the high performance and low latency of the technology, we see a similar parallel to the days of DAS. Most PCIe Flash comes in the form of a card that plugs into the server s PCIe slot and remains captive to that server. Just like DAS, the device is not shared, and works independently as dedicated storage to a given application or server. As PCIe Flash capacities continue to grow (already at 4.8TB and growing for half-height half-length cards) the need to share these devices becomes more important. Likewise, SAN features such as replication and caching should be implemented to maximize the investment in this new storage tier. A recent report by Wikibon.org has caused quite a stir in the enterprise storage community as Wikibon.org suggests that traditional enterprise storage will decline as a new technology called Server SAN, comprised of commodity servers and PCIe storage, will take hold due to ease of management, higher performance and ability to scale-out cost effectively. Highlighted below is the Wikibon.org analysis: Traditional Enterprise Storage, Hyperscale SAN & Enterprise Server SAN Revenue Projections 2012-2027 Regardless of whether you believe the ramp of the change or the timing of it, Wikibon.org s assertion that enterprises will consider Server SANs is intriguing. But there are gaps in today s technology that are holding back this data center transformation. The gaps include shared flash storage, high availability, caching and cluster-wide volume management. HGST Virident Solutions 2.0 Companies like HGST are bringing SAN capabilities server-side to deliver shared storage, high availability, caching. HGST s Virident Solutions address the limitations of DAS PCIe Flash with an open architecture software for servers that bring the benefits of a SAN to previously stand-alone devices. HGST calls this a Flash Fabric. The HGST Virident Solutions Suite includes HGST Virident Share for remote access of a FlashMAX SSD storage device, HGST Virident HA synchronous replication, HGST Virident ClusterCache acceleration for databases as well as HGST Virident Space for cluster-wide flash volume management. All of these applications work in conjunction with vfas, HGST s Virident Flash Aware RAID with Adaptive Scheduling operating system that runs on the FlashMAX family of PCIe SSDs. 1
Server-side flash has distinct advantages over flash deployments using SAN, NAS or iscsi. By placing flash storage directly into the server and using intelligent software, like vfas, the number of protocol layers and hardware hops is significantly reduced, thereby lowering latency and improving response times from milliseconds to microseconds. Shared Flash Storage As we have described, today s server-side flash is available as a direct attached storage device only. This means that the flash device is dedicated to a single server and often grossly under utilized. The effective cost/gb for a dedicated card is often high and gets higher the lower the utilization. HGST s breakthrough technology, Share, uses Infiniband or Ethernet to provide a server-side. enefits include remote access from servers that do not have SSDs, along with increased resource utilization through sharing. To consider the financial benefits of shared storage using Share, we calculate the desired flash capacity that will be allocated per server. Then we purchase the total number of SSDs required. In the case of current flash competitors, this is at least one SSD per server. In HGST s case with Share, only a single card is required, as it will be shared across the Server SAN network. Since a single FlashMAX SSD can deliver more than 350,000 random 4K Random Reads, and hold 4.8TB of MLC flash, we need only two devices to service a network of 6 servers requiring 1.6TB of flash each. Server-Side Replication Since today s flash devices are direct attached and do not interact with each other, they represent a single point of failure. This means a single device failure can bring down an application. Some software products like VMware VSAN address this issue with server-based software for replication. But VSAN is VMware-only. HGST delivers flash fabric high availability with device-affinity meaning that the replication is associated with the specific volumes on the SSD for granular replication and recovery. HA has a simple approach to replication enabling highly granular Recovery Point Objectives for database availability. The solution also supports multiple FlashMAX devices and replicated stores in a single server. 2
HA works with multiple cluster management services to provide a high availability storage subsystem. In the case of MySQL the DRBD, Distributed Replicated Block Device, can be configured to provide synchronous replication of a device. But since it is unaware of the underlying storage device, and has not been optimized for the exceedingly high IOPS and bandwidth of the HGST FlashMAX SSD, it is simply unable to provide the same performance as HA, even when coupled with the same FlashMAX device. HA Simple Integration For Oracle or MySQL, HA integrates with the widely available Pacemaker and Corosync stack. Since HA works in conjunction with the vfas layer, it is possible to provide the highest performance replication. Caching In the case where an existing storage network is in place and applications still need lower latency and faster response times, caching delivers performance and durability unequalled in the market. Rather than buy more spinning disks or SSDs in the storage array, deploying caching server-side can have a significant ROI. All flash deployments will always be faster than systems using flash merely as a cache, but even so, there are good reasons for going with HGST Virident ClusterCache over an all-flash array solution. First the total amount of storage array flash needed may be uneconomical for the datasets in mind. Secondly, for large datasets, it may not be possible to install enough capacity. Finally, even for large databases, often only portions are actively used. A cache allows for automatic tiering of hot data to flash without user intervention. Optimized Flash Endurance ClusterCache builds the cache directly into FlashMAX accessing and managing flash memory directly while eliminating double mapping of block addresses which can reduce endurance when using generic software caches. Clustered Caching for Oracle RAC Oracle Real Application Clusters (RAC) environments cannot benefit from PCIe Flash since RAC requires shared storage. With ClusterCache and Share, Oracle RAC can take full advantage of the high performance and low latency of FlashMAX PCIe SSDs. ClusterCache with Share helps minimize the cost of high availability Oracle RAC clusters while providing the highest available flash performance and reducing the need for expensive SAN storage solutions. It effectively converts the HGST FlashMAX SSDs into a shared storage infrastructure capable of full integration with Oracle RAC and managed by Oracle Automatic Storage Management. 3
ClusterCache with Share delivers significant performance advantages compared to a configuration where no PCIe Flash is resident. Using Oracle Enterprise Manager it is possible to see 10X improvement in throughput and more than 47x increase in IOPs as certified by the Enterprise Strategy Group and illustrated here. Volume Management The biggest challenge to making the successful is enabling share flash storage to be concatenated into a single large volume for shared database applications like Oracle RAC and to allow large volumes to be fully mirrored for resiliency. In addition to these critical functions, a volume manager should enable that large concatenated volume to be carved up and delivered to application servers based upon the amount of flash required, just like a traditional SAN. This function is missing from today s flash devices. HGST has introduced HGST Virident Space to address this need. Space is a server-side flash volume manager that can cluster up to 128 servers, 16 FlashMAX devices and concatenate these into a 38.4TB fully mirrored flash volume. It allows all hosts in the cluster to see all volumes. Space allows the cluster to grow dynamically by adding a new FlashMAX SSD into a server or by adding a new server with an SSD. Failover is automatic and transparent to any application using the cluster. The devices can move between servers for easy upgrades. HGST Virident Space Highlights Up to 128 hosts and 16 FlashMAX in a cluster Using 4.8TB FlashMAX II = 38.4TB logical volume and 2-way mirroring between each device Concatenate all devices into a pool All hosts in clusters can see all volumes Grow cluster dynamically Ethernet or Infiniband Financial Benefits of Clustering Traditional direct-attached islands of flash might have some devices that are 10% utilized and others that are 90% utilized. For servers that have devices fully utilized, you have to buy more even though you have unused capacity on other machines. Utilizatiton 50% 90% 4.8 TB FlashMAX II Device $24,500.00 $24,500.00 Utilization Cost $12,250.00 $22,050.00 Unused Capacity Cost $12,250.00 $2,450.00 Clustering Saves Money In a 16-node configuration with 8 nodes at 10% utilization and 8 nodes at 90% utilization, you could be wasting as much as $100,000 on unused flash. # of Devices 8 8 Total Unused Cost $98,000.00 $19,600.00 Flash for KVM For Kernel-based Virtual Machine (KVM) environments, Space turns the KVM environment into a virtual SAN where a pool of Flash may be used for application acceleration. This brings VMware VSAN-like capabilities to KVM environments. 4
Flash for GFS Use Space in conjunction with Ret Hat Global File System to create an all-flash shared file system with blazing performance and granular volume management. Server Consolidation for MySQL HGST Virident Space provides a unique solution for server consolidation in MySQL environments. Using Space, you can set replication from multiple masters to multiple slaves using unique space volumes. A separate Stand-by Server can be added to the cluster. It can be powered off to save opex until it is needed. When it is needed, the Stand-by Server can mount any of the replicated volumes from the Space pool and be used for recovery. Using a cluster of 5 masters and a stand-by, a server consolidation of 37% or more may be achieved. Software for the HGST Virident Solutions comprise a complete software suite that delivers shared storage and replication across the server-side network with breakthrough capabilities unavailable from other suppliers. Using the Server SAN, this software includes Share for shared storage and remote access, HA for synchronous replication along with ClusterCache software that accelerates application performance for existing networked storage. With the introduction of Space, revolutionary new capabilities for can be achieved. Never before has there been technology for PCIe SSDs that allows for clustering and aggregation of Flash that resembles a SAN Array Volume Manager. By layering this volume management capability with sharing and high availability, the next generation performance tier of storage is emerging. Collectively, this set of software provides a broad foundation for building a. Regardless of your opinion on the Wikibon.org forecast, there is sufficient evidence in markets like public cloud and hyperscale that illustrate the rapid adoption of PCIe SSDs and commodity servers with lower power consumption, easier management and faster time to market. HGST s is leading this data center transformation and the deployment of Server SANs. About HGST HGST, a Western Digital company (NASDAQ: WDC), develops innovative, advanced hard disk drives, enterprise-class solid state drives, external storage solutions and services used to store, preserve and manage the world s most valued data. HGST addresses customers rapidly changing storage needs by delivering intelligent storage devices that tightly integrate hardware and software to maximize solution performance. Founded by the pioneers of hard drives, HGST provides high-value storage for a broad range of market segments, including Enterprise, Cloud, Datacenter, Mobile Computing, Consumer Electronics and Personal Storage. HGST was established in 2003 and maintains its U.S. headquarters in San Jose, California. To find out more about HGST enterprise-class HDD, SSD and SW solutions, please visit www.hgst.com. 2014 HGST, Inc. 403 Yerba Buena Road, San Jose, CA 95135 USA. Produced in the United States 8/14. All rights reserved. Virident and FlashMax are registered trademarks of HGST, Inc. and its affiliates in the United States and/or other countries. HGST trademarks are intended and authorized for use only in countries and jurisdictions in which HGST has obtained the rights to use, market and advertise the brand. Contact HGST for additional information. HGST shall not be liable to third parties for unauthorized use of this document or unauthorized use of its trademarks. All other trademarks are the property of their respective owners. References in this publication to HGST s products, programs, or services do not imply that HGST intends to make these available in all countries in which it operates. Product specifications provided are sample specifications and do not constitute a warranty. Information is true as of the date of publication and is subject to change. Actual specifications for unique part numbers may vary. This paper contains forward-looking statements, including statements relating to expected demand for storage products and the growth in data generation. These forward-looking statements are subject to risks and uncertainties that could cause actual results to differ materially from those expressed in the forward-looking statements, including changes in markets, demand, global economic conditions, the development of potentially competing technologies, and other risks and uncertainties listed in Western Digital s recent SEC filings, to which your attention is directed. Readers are cautioned not to place undue reliance on these forward-looking statements, which speak on as of the date hereof, and HGST/WD undertakes no obligation to update these forward-looking statements to reflect subsequent events or circumstances. Please visit the Support section of our website, www.hgst.com/support, for additional information on product specifications. Photographs may show design models. WPFLASH14EN-01 5