Shared DAS Concept - An Alternative to SAN?

Traditionally, if the server’s own storage subsystem is not enough, the choice is limited to DAS (directly connected disk shelves) and network block or file storages. But recently, another very promising concept has been added to these options - Shared DAS. Why is she good?



First, let's go over the pros and cons of the existing options.

Let's start with the simplest, that is, Directed Attached Storage (DAS), a typical example of which are disk shelves connected via SAS to the controller in the server. A typical DAS list of benefits is as follows:

• High performance
• Low cost
• Simple operation
• Data Protection with RAID Controller

Disadvantages:

• Inability to scale to multiple servers
• Inability to build high availability solutions

Everything is very obvious. If the requirements for continuity of operation are high, then instead of DAS we already use a shared storage system with direct connection ports, which provides several controllers. Shared Storage Advantages of this approach:

• Cost is still lower than full SAN
• Ability to expand disk space for multiple servers
• Simple operation
• Fault tolerance with full duplication of components

Minuses:

• Modest performance for entry-level systems
• Link to proprietary solutions and components
• Scalability is significantly limited by the number of ports in the system.
• Relatively high cost
• The need to license many software features

Finally, networked storage in the form of a Storage Area Network (SAN) built on FC or iSCSI. SAN Advantages:

• Fault Tolerance and Service Continuity
• Extensive data management capabilities
• Good scalability

Disadvantages:

• High price
• High configuration and management complexity

Different ideas have long circulated in the industry on how to make a highly accessible solution suitable not only for large companies, but also for the SMB segment. The number of such enterprises on the planet is hundreds of thousands, a huge and empty market.
That solution was Shared DAS, shared disk storage. From disk shelves, this concept has absorbed low cost, high performance and ease of management, while high reliability and continuous availability have been taken from network solutions.
The development of SAS helped a lot here, which, from simple transport, became a developed network protocol with the ability to switch, zon, and to ensure data integrity, the T10 information security model was adopted. Shareddas




The basis was the use of SAS as a switching environment and the use of self-configuring port expanders (SAS expanders). It is SAS that provides completely duplicated access to drives and the ability to use extremely heterogeneous drives in the same system: a mixed storage medium can simultaneously include SATA and SAS disks, as well as SSD drives, and SAS devices can be of different generations (3G, 6G , 12G). Due to the SAS switches, which, in fact, are the same expanders in the external design, all this wealth can be combined into pools, organizing tiered storage with the separation of data according to processing speed requirements.
A separate additional advantage of SAS is the extremely low latency of data transfer at the interface level - if in the case of hard drives, the response time of which was measured in milliseconds, this could be neglected, then in the case of SSDs with a response of tens of microseconds, this moment came to the fore. SAS switched environment allows for multi-level data storage with separation of them into “hot”, “warm” and “cold” according to their access speed requirements in the most natural way, by scaling a single storage network and operating with drives at a logical level, without the need to separately scale drives for each group of requirements. scaling




Excellent SAS system scalability is provided by cascading switches. So, LSI SAS6160 provides a cascading depth of 6, the total number of devices in the cascade can reach up to 1000, while the switches are configured out-of-band relative to the data transmission medium - over TCP / IP, Telnet. The current state of equipment development allows you to create complex systems, including dozens of servers and hundreds of drives. Access through multiple paths




As for the fault tolerance of the infrastructure built on SAS, it is ensured by the full duplication of all routes, including ports directly on the drives. An important note should be made here: the use of SATA drives, although it is possible in such an infrastructure with the help of interpozers (multiplexers that provide access to SATA in two ways), will nevertheless reduce the overall reliability, since it is the interposer that will become the very point of failure. The switches are capable of forming full-fledged independent channels for accessing disks and, in the event of a failure of one of the paths, automatically redirect the load to another channel. The role of the OS and control applications in these conditions is to provide failover transparent to the user.

Since we are talking about operating systems, it is worth noting that the solutions within the framework of the SharedDAS concept have long been available on the basis of ZFS, but the configuration of such systems required rather deep specific knowledge, and some features of ZFS can be taken into account only at the design level system. Therefore, the market of fault-tolerant solutions became truly widespread with the release of Windows Server 2012, which included the Storage Spaces component, which made it possible to virtualize data storage at the OS level. Storage Spaces Feature Example




Storage Spaces introduced spaces called Storage Spaces - virtual storages based on groups of physical media grouped in Pools. This approach made it possible to separate the management of storage media from the management of data warehouses. The main value of this component is that it supports the sharing of SAS devices, which means that in a server cluster running Windows Server 2012, the common Storage Spaces space will be available immediately to all nodes in the cluster and can be used as part of the Cluster Shared Volumes (CSV) volumes .

Storage Spaces provides:

• Creation of one or several storage pools, including hybrid (from HDD and SSD combinations), which can be dynamically expanded by adding drives
• Constant availability using the clustering component, available only for SAS storage (FC and iSCSI are not supported), pools are transferred between the cluster nodes as needed
• Balancing CSV membership to different cluster nodes depending on load
• Different levels of data protection, such as backup drives (Hot Spares), background checking, error correction mechanism, different RAID levels (Simple, Mirror, Parity, Dual Parity)
• Optimize Disk Usage (Thin Provisioning)
• Multi-tier storage of hot and cold data on hybrid pools
• SSD write caching

And most importantly - everything is done extremely simply and clearly.

As for us, we already talked about the ETegro Fastor FS200 G3 - a fault-tolerant storage server built on this concept. In addition to it, on the basis of the SharedDAS concept, we already offer ETegro Hyperion RS420 G4 - a Cluster-in-a-box class server, about which there will be separate material for building fault-tolerant solutions of various types and a number of solutions based on disk shelves.

If anyone thought that we were sold to Microsoft, then in vain - we will soon publish a way to raise a similar high-availability cluster on Linux.