HP's All-flash Array and 10 More Big Changes to 3PAR Storage Systems (Part 1)

    Today at HP there was a big announcement in the storage department, I want to share this information. Models of mid-range storage systems optimized for working with flash drives were announced - the HP 3PAR StoreServ 7450, and the functionality of the current

    7000/10000 systems was expanded. HP 3PAR StoreServ 7450


    Almost any storage system currently available on the market supports SSD drives.
    Traditionally, the requirements of high performance from the disk subsystem were achieved by adding a large number of rotating spindles, usually with minimal utilization of their volume (Fig1).
    The use of SSD drives has reduced the number of rotating spindles, while increasing the performance of data arrays.


    Fig. 1 Achieving high performance in traditional arrays and modern arrays



    At a time when the market is dominated by traditional arrays with an almost 20-year architecture or ultra-expensive all-flash startups, HP offers an array with a completely different architecture, specially designed for simultaneous work with various types of data. The key features of the new 3PAR 7450 array are that controllers are not a performance bottleneck. This ensures almost linear performance growth as new drives are added.

    Technologically, these arrays are similar to the current 3PAR 7400 with improved cache and interconnect characteristics between controllers. Like the HP 3PAR 7400 model, the new 7450 can expand to an array of 4 fully-connected controllers with write-cache mirrors, and the uniqueness of the array is that each array controller is active for all volumes - which gives fair load balancing between all nodes and all array disks. This unique technology allows you to maintain high performance even if one controller fails.
    The 7450 arrays have a large cache volume - 64GB for a dual-controller system and 128GB for 4 controllers, while the write cache has its own channels, so the array operating system cannot “eat out” the write cache for its needs and it is fully accessible for input operations / output.
    Due to such a large cache level, the use of ASIC chips and eight-core Intel E5-2470 2.3GHz processors, the array can overclock to 520,000 IOPs using only SSD disks (4k random read), which is important when the delay is less than 1ms. And to get such large numbers, only a few shelves with SSD carriers are enough. Large IOPS require large array throughput. Thanks to the use of ASIC chips, the central processor of each controller is very unloaded, this allows you to get very high throughput rates - up to 5.2 GB / s. Thus, the "raw" computing power of each controller has increased by more than 55%.
    To cope with such heavy loads and support clustering, bus bandwidth between controllers has been increased.
    Because Since most traditional disk arrays are built on the architecture of 20 years ago, the use of SSD disks in them does not bring significant performance, but only leads to a short-term improvement in reading performance. The controllers of such arrays have significant limitations on the number of supported SSD drives.
    The architecture features of the 3PAR array are such that they provide high performance even when adding a large number of SSDs (120 SSDs for the 3PAR 7200 model, 240 SSDs for the 3PAR 7400/7450 models).
    To assess the wear of SSD drives, the advanced Flash Wear Gauge mechanism is used, which will inform you in time about possible problems with the media.

    Adaptive cache

    For the 3PAR 7450 array, the microcode for working with the controller cache was redesigned.
    The minimum page size stored in the cache of most storage systems is 16KB. The 3PAR array has a “smart” cache - for each type of disk, depending on its size and quantity, a specific area of ​​the write cache is allocated. Further - if sequential and random accesses go to this type of disks in parallel, then the pages of this part of the cache are adapted to request types. For example, with several sequential read operations, the cache immediately reads several blocks of 16K each, thus reducing the load on the media and reducing the response time. As soon as consecutive calls ceased, only the current pages remain in the cache.


    Fig. 2 Adaptive cache in the 3PAR 7450 storage system

    Consider an example where a host requests 4KB read. Traditional cache-to-disk arrays are characterized by 16KB block accesses. This is because there is a delay in rotating disks, so a larger block is immediately placed in the cache in order to predict further accesses to neighboring blocks. This approach will increase the speed of accessing the array, but load the data cache heavily. SSD disks are characterized by ten times shorter response time, therefore, the 3PAR 7450 arrays use an adaptive way of working - the array cache requests a block from SSD carriers that is almost equal in size to the I / O block requested from the host, which loads the cache and SSD disks less.


    Fig. 3 Adaptive Read Cache in 3PAR 7450 Array

    Recording requests complicate the situation. It is known that SSD media processes recording in 4/8 KB blocks from the host, but overwriting the SSD cells of the media is done in blocks of 64KB, for early models of SSD drives - 512KB (rewriting cycles for SSDs are always limited). The approach of traditional arrays is that 16KB pages are stored in the cache and blocks of the same size are written to SSD media, even if the server sends blocks at 4 / 8KB, i.e. an incomplete page will be stored in the cache until it is full, only after that it will be written to SSD media. But more often than not, an incomplete page will be reset on SSD media, as in arrays, policies are applied to force the cache to be flushed to disks after a certain timeout has been reached to save data (even if only 4KB is written to the controller’s cache, the 16 KB block is reset, of which 12KB will be filled with zeros). In this approach, the efficiency of using SSD media and their service life is greatly reduced. This behavior is typical for most arrays of the traditional type.
    The approach used in 3PAR storage systems - smaller blocks (up to 4KB) are flushed from the cache to disks, this approach has several advantages:
    1. SSD media wear is reduced because only small data blocks will be written
    2. The mechanisms for finding free space on an SSD are improved after filling the media: an array is easier to find a free block in 4KB than a block in 16KB
    3. Based on p. 2, the utilization of the SSD media is improved - only blocks with data are stored, blocks filled with zeros are excluded
    4. For RAID1 volumes, the delays and penalties associated with the sequence of read-modify-write requests are reduced



    Fig. 4 How

    does the 3PAR 7450 cache write? If the host receives write requests in large blocks (for example, 128KB), the 3PAR array cache splits them into smaller blocks and shares the load between the media. In this approach, each SSD carrier wears out much less. In addition, the controller can simultaneously process requests of different types: sequential and random. Each type of request is simultaneously processed by ASIC chips, thereby reducing the queue and increasing the speed of the array in mixed loads. Arrays of a traditional type with mixed loads are characterized by large queues of requests.


    Fig. 5 Principle of operation of the 3PAR array cache in mixed loads

    What are the advantages of such a storage system, along with the many projects of Flash-arrays that have appeared in the near future?
    • Firstly, the 3PAR 7450 array provides high performance comparable to flash arrays. It uses SSD drives, which are much cheaper than Flash-modules. Traditional storage systems, due to their outdated architecture, cannot achieve high performance.
    • Secondly, the 3PAR array supports Disaster recovery, which manufacturers of Flash arrays cannot offer yet.
    • Thirdly, Flash-arrays, as well as arrays of traditional architecture, cannot be expanded to 4 controllers with automatic load balancing between them.
    • Fourth, the Flash array itself is usually weakly connected with other storage systems, which reduces the area for its use, while 3PAR allows you to store all types of data on one storage system, with one management interface and dynamic data movement between levels .
    • Fifthly, 3PAR arrays can be federated from several systems, with load balancing between arrays, which is not yet available in both Flash solutions and arrays of traditional architecture.


    Disaster Recovery Support

    In addition to performance improvements, 3PAR arrays have increased the ability to create fault-tolerant solutions. In particular, the work of Peer Persistence technology aimed at collaboration with VMware solutions has improved. There are several improvements:
    1. The load balancing features between the two sites were implemented in conjunction with Storage vMotion. Now, VMware's automatic load balancing is possible not only within the array between all controllers, but also between multiple arrays.
    2. The High Availability feature was implemented for virtual machines between multiple data centers.
    3. Synchronous replication now supports “never disappearing LUN presentation” technology:

    The volume is replicated between arrays and fault-tolerant paths presented to the host are created for it. Disk arrays located at different sites will be interconnected by replication implemented using 3PAR Remote Copy. Due to the fact that both sites are active, replication will be two-way. 3PAR Peer Persistence technology provides vMSC cluster nodes with uniform access to two storage systems at data center sites, while each logical drive (LUN) of each 3Par storage system has a synchronous copy on another site (as shown in the diagram below). This copy is presented to the VMware virtual environment as a passive path to the primary LUN, as it has the same WWN identifier (a unique identifier for the device in SANs). In the event of a failure of the primary LUN, its backup remains visible to the hosts, thus By enabling the hypervisor to migrate virtual machines using conventional vMotion, High-Availability, or Fault-Tollerance technologies. Switching between sites is performed by a quorum, i.e. This is a fully automatic operation.
    Most traditional technologies (such as, for example, the automated solution VMware Site Recovery Manager) in the event of a storage system failure can transfer the entire virtual environment to a backup site, but this will require a reboot of the virtual machines. Thanks to the Peer Persistence technology described above, a storage system is available for a virtual cluster even when the disk array of one of the sites is out of order. At the same time, switching to the backup site occurs without the use of third-party mechanisms, which makes it possible to transfer virtual machines to another site without stopping them and, at the same time, does not complicate infrastructure management.


    Fig. 6 Disaster Recovery Technology in VMware

    Application clustering software is still available for physical machines, working in conjunction with 3PAR arrays - HP Cluster Extension for Linux and Windows and MetroCluster for HP-UX.

    3PAR StoreServ 7000/10000 Array Encryption System

    Since June, customers have the opportunity to order in addition to the standard 3PAR 7000/10000 arrays arrays with data encryption. The difference between these arrays is the availability of special SED (Self-Encrypted Disk) disks, each of which has an encryption chip. Encryption protocols AES-128 / RSA-2048 / SHA-256 / RNG FIPS-186 / SP 800-90DRBG are supported. Due to the low latency of the chip, the speed of the SED disk is practically no different from the speed of a regular disk. Upon initialization, the encryption key is recognized by the Local Key Manager, which is part of the array operating system and runs on each controller. During the first initialization, the array will be asked to save a copy of the key. In the event of a power failure on the disks or array, the data is blocked and access to the data can be restored by re-entering the encryption key in the Local Key Manager console. If an IT specialist leaves the organization, then you can change the key. When migrating to an array without the encryption function, the data will be decrypted, while migrating to the opposite side, it will be encrypted. There are important nuances:
    • Encryption is activated by a license; one license per array is acquired.
    • You cannot mix regular and SED disks in an array with the encryption function activated.
    • If the encryption function is not activated, then you can mix disks in an array, but the cost of SED disks is approximately 5% higher.
    • After the encryption function is activated, all SED disks must be reinitialized, i.e. data on disks will be unavailable.

    Several types of SED drives will be available for encryption - 400GB MLC SSD, 450GB 10k, 900GB 10k, 1TB 7.2k 2.5 ”.


    Fig. 7 Scheme of data encryption in the 3PAR array

    Enabling the encryption function is very fast, in several parallel streams, for example, on a system with 160 disks, encryption is applied in ~ 30 sec.

    What environments is the HP 3PAR 7450 array designed for?
    Firstly, these are OLTP / WEB 2.0 technologies, for which response time and array speed are crucial, and for the enterprise this directly affects the profit.
    Secondly, these are VDI technologies that have become widely demanded, which are characterized by the presence of an avalanche-like growth in accessing the array within a very short period of time (boot storm). Using the 3PAR 7450 fast array will allow you to squeeze the maximum performance out of SSD drives (up to 10,000 IOPS from the media), thus reducing their number.


    Fig. 9 Pattern of typical array calls for VDI

    Thirdly, this is business analytics. Quick analysis of large amounts of data - quick response to changes and high profits.

    What announcements are worth mentioning:
    1. You can order MLC SSD disks 400GB - for those cases when large SSD volumes are required, NL SAS disks 1TB 7.2k 2.5 "- for those who need capacious storage in the form factor 2.5" (for 3PAR 7000), SAS disks 600GB 10k 2.5 ".
    2. Host Priority Maintenance Technology (QoS).
    3. Announced support for OpenVMS, Ubuntu.
    4. Support for OpenStack interfaces and drivers has been announced, the 3PAR array becomes an object data warehouse and is even easier to integrate into the “clouds”.
    5. FCoE connectivity is now supported on devices with multiple hopes.

    UPD: A protocol for testing 3PAR Peer Persistence functionality appeared on the VMware website . Pay particular attention to the table of virtual machine testing scripts in the cluster.
    List of references:
    1. HP 3PAR 7450 Quickspecs
    2. HP 3PAR 7000 Quickspecs
    3. HP 3PAR 10000 Quickspecs

    Also popular now: