Introduction to the IBM Elastic Storage Server
This chapter introduces the IBM Elastic Storage Server solution, its characteristics, and where it fits in the business environments.
This chapter also describes some of the software and hardware characteristics of the Elastic Storage Server, the software Redundant Array of Independent Disks (RAID), and the building block concepts of the solution.
The following sections are presented in this chapter:
1.1 Elastic Storage Server
The IBM Elastic Storage Server is software-defined storage that combines IBM Spectrum Scale (formerly GPFS), IBM software RAID, and performance of IBM POWER8® architecture.
The building block-based solution of the IBM Elastic Storage Server delivers high performance, high availability, and scalable functionalities on clustered file systems for a wide variety of demanding applications.
IBM Spectrum Scale is required to access data on the Elastic Storage Server common repository infrastructure of shared file system through native clients. IBM Spectrum Scale’s new capabilities include protocol nodes to allow object access data. They use OpenStack Swift or file access through Server Message Block (SMB) or Network File System (NFS).
1.2 Software RAID
The IBM Spectrum Scale RAID software that is used in the Elastic Storage Server solution runs on standard serial-attached SCSI (SAS) disks in just a bunch of disks (JBOD) arrays. The solution does not require or use any kind of external RAID controller or acceleration. The RAID functions are handled by the software. Use of SAS drives allows greater cost reduction. The solid-state drives (SSDs) option is also available when more performance is needed.
IBM Spectrum Scale RAID supports multiple RAID codes and distributes client data, redundancy information, and spare space across the disks. This approach ensures that if there is a physical disk loss, or even a group of physical disk losses, data availability is unchanged.
Spectrum Scale RAID implements an end-to-end checksum to detect and report faults, read or write errors, and other integrity problems of traditional RAID.
1.2.1 RAID codes
Spectrum Scale RAID in the Elastic Storage Server supports different data protection algorithms that can detect and correct up to three concurrent faults.
The options for RAID configuration are eight stripes of data plus 2 or 3 parity stripes. You can use Reed-Solomon codes or one stripe of data plus 2 or 3 replica stripes. The data plus parity or replica stripes, which are called tracks, are illustrated in Figure 1-1 on page 3.
Figure 1-1 RAID tracks
1.2.2 End-to-end checksum
IBM Spectrum Scale software on the client is used to access data on the Elastic Storage Server. This software is aware that the Spectrum Scale file system is based on Spectrum Scale RAID Network Shared Disks. During a write operation an 8-bytes checksum is calculated, appended to the data, and sent over the network to the Spectrum Scale RAID server. The checksum is verified. Then, Spectrum Scale RAID writes the data along with its checksum on the disks and logs the version number in its metadata.
When a read operation is requested, Spectrum Scale RAID verifies checksum and version on its metadata. If it is OK, it sends the data to the client. If it is not OK, the data is rebuilt based on parity or replication. Then, the data is sent to the client along with newly generated checksum.
The end-to-end checksum feature can prevent and correct silent disk errors or missing disk writes.
1.2.3 Declustered RAID
Spectrum Scale RAID implements its own data and spare disk layout scheme, which reduces the overhead to clients during a recovery from disk failure. To accomplish this, it does not leave all spare space in a single disk. Instead, it spreads or declusters user data, redundancy information, and spare space across all the disks of the array. Figure 1-2 on page 4 compares conventional 1+1 RAID layout with a declustered array.
For example, consider seven stripes of data on each disk. Figure 1-2 on page 4 shows the left three arrays of two disks in a replicated 1+1 configuration and a spare. On the left, you can see the data stripes that spread over all seven disks of the declustered array.
Figure 1-2 Declustered array versus 1+1 array
In Figure 1-2, notice that on the Elastic Storage Server from v3.5 and even on GL6 models, only one declustered array is used. Figure 1-2 shows 348 HDD, which is a simplified model.
If one disk fails on the traditional 1+1, all data from the remaining disks of the array must be replicated to the spare disk. On the declustered array, the replication occurs on spare space of all the remaining disks. This approach can decrease the rebuild impact by as much as four times. Figure 1-3 shows a simplified model.
Figure 1-3 Array rebuild operation
Consider the case of an Elastic Storage Server that uses RAID 8+2 or two-way replication (1+2). If one disk is lost, the rebuild operation starts with low priority with even lower impact for the clients. With this array configuration, two concurrent disk losses must occur before the system treats the rebuild as critical to be run on high priority. Using 8+3 RAIDs or three-way replication (1+3), the rebuild operation becomes critical only when three concurrent disk losses occur in the same declustered array.
1.3 Building blocks
The building block is the minimum configuration of an Elastic Storage Server and is also the unit of expansion of the solution. If more space is needed, either a new building block is added to the solution, or it is expanded to a higher model.
Each building block consists of two IBM POWER8 servers and a specific number of storage enclosures (1, 2, 4, or 6), depending on the model.
The POWER8 servers are model S822L and the storage enclosures can be one of the following types:
•DCS3700 (4U 60 drives) units for models GL2, GL4, and GL6
•EXP24S SFF Gen2-bay drawer (2U 24 drives) for models GS1, GS2, GS4, or GS6.
The GS models can use 2.5" 10 K rpm HDD or 2.5" SSD. GL models can use 3.5" NL-SAS HDDs. Models GL and GS can be mixed to achieve specific needs for client applications that use the storage. For example, SSDs for metadata and NL-SAS for data storage.
Elastic Storage Server V5.3 introduced two new hybrid models to the ESS family: GH models. ESS Hybrid models place spinning disks and flash storage in a single ESS node, giving better storage density, and a smaller footprint at a lower cost. GH14 has one 2U drawer with 24 Solid State Drives (SSD) combined with four 5U drawers with a 7200 rpm spinning disk. The GH2R has two 2U drawers with four 5U drawers.
Like the GS models, the SSDs are either 3.84 TB or 15.3 TB capacities. The 5U drawers are similar to those in the GL models, either 4 TB, 8 TB, or 10 TB capacities.
A new Enterprise Slim Rack (S42) is now available to hold these. The S42 is available for all ESS orders, including the GS, GL, and new GH models. You can refer to the following announcement letter for more details: https://ibm.biz/BdYkqC.
|
Model upgrades: For clients with existing Power Systems servers, nondisruptive upgrades are introduced with Elastic Storage Server V5.3. With nondisruptive upgrades, the original data is preserved and accessible in place. Meanwhile, the added capacity of the new storage drawers is integrated into the original capacity without interruption. The existing data is rebalanced across both the new and old storage drawers, again without interruption to the Elastic Storage Server cluster. The new storage capacity is immediately available for use.
|
1.4 Value added
IBM Spectrum Scale RAID and support from IBM for the solution, contribute to the added value, including a tuned solution from the IBM services team.
IBM engineering and testing teams created a solution that provides greater performance and data availability. They included these design factors in the solution:
•adapter placement on the servers
•number of adapters
•number of disks on the drawers
•cabling
•number of drawers to the software versions
•the way data is placed on disks
Elastic Storage Server offers scalability from 40 TB to hundreds of petabytes. It supports 10-Gigabit Ethernet (10 GbE), 40-Gigabit Ethernet (40 GbE), 100-Gigabit Ethernet (100 GbE), and 100 Gb/s EDR InfiniBand.
Big data requires easy storage growth. The IBM Elastic Storage Server building block approach meets that requirement. Adding more storage servers adds to the overall capacity, bandwidth, and performance with a single name space.
By leveraging JBODs and IBM Power Systems servers, the Elastic Storage Server provides price/performance storage for analytics, genomics, video streaming, machine learning, metadata services, artificial intelligence, technical computing, and cloud computing environments.
The Elastic Storage Server’s modern declustered RAID technology can recover from multiple disk failures in minutes, versus hours and days in older technology. As a result, the solution delivers predictable performance and data protection. 8+2 and 8+3 RAID protection and platter-to-client data protection are included with the Elastic Storage Server.
Various tools and scripts are used by the services team during the delivery. These items ensure that every piece of the solution integrates together and customer requirements are met.
For more information about the Elastic Storage Server, refer to the following websites:
https://ibm.biz/BdYkqW
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.