Frequently asked questions about SDS software-defined storage

What is the difference between software-defined storage and traditional storage arrays?

1, different scalability:

Traditional disk array: it is mainly composed of "control cabinet" and "hard disk expansion cabinet". The "control cabinet" is connected in series with the "hard disk expansion cabinet" and "hard disk expansion cabinet" through SAS cables, and will continue to expand in the future.

The moment you choose the "control cabinet" model, you have determined the maximum scalability (performance &; Capacity).

The "control cabinet" (or main control cabinet) will have two dedicated main/standby controllers (similar to servers, with special CPU and memory, etc. ), they are responsible for receiving IO from the server, and dispatching the robotic arms in multiple disks in the "control cabinet" and "hard disk expansion cabinet" to finally unload the data, which is equivalent to the brain center of the whole set of storage.

Therefore, the processing capacity of the controller of the "control cabinet" determines the maximum IOPS (performance) of the whole set of storage, the number of "hard disk expansion cabinets" that can be connected in series, and the type and number of disks that can be controlled. (capacity)

Software-defined storage: It is to deploy distributed storage software on N X86 servers, and use local disks inserted on N servers to store data. NameNode of distributed storage (similar to brain node, which is responsible for calculating how data is distributed) receives IO from the server that needs to write data, and through software calculation, tells the server that needs to write data to finally put the data into a DataNode (the node that finally writes data). Computing power is mainly limited by NameNode, and IOPS is mainly limited by the number of DataNode, the number of disks and the type of disks. Once the performance/capacity is insufficient, you can scale out by adding NameNode or DataNode, which is theoretically infinite.

2. Different compatibility:

Traditional Disk Array: Once the brand is selected, basically the subsequent expansion can only be hacked. Because disk arrays between different brands cannot be mixed. For example, if you have 48TB of Dell storage and 1 of HP storage, you can't mix all the raw capacity into a 96TB area and then extract logical volumes from it.

Sometimes, different series of the same brand may not be compatible.

Once the brand model is determined, subsequent expansion can only be kidnapped by manufacturers, and there is no bargaining power. Moreover, once the product is updated, the series is discontinued due to the manufacturer's reasons, and it is necessary to buy a new series and it is incompatible with the series storage being used, so it can only be forced to do data migration.

In order to solve the above-mentioned cross-brand or cross-series mixed storage resources of the same brand, it is necessary to purchase additional hardware and increase additional expenses.

Software-Defined Storage: Because they are all deployed on the operating system of X86 servers, no matter what brand of servers are used, the operating system is the standard Linux or Windows operating system, so there is no problem in compatibility, and it can be compatible across brands of servers.

3. Different usability:

Traditional Disk Array: The basic disk arrays are all commercial products, and each manufacturer basically has a graphical management interface, so the daily operation and maintenance can be basically completed by clicking the graphical interface. And once there is a problem, you can also file a case for the assistance of the manufacturer. After all, what is stored is data, which is the core productivity of an enterprise. Once the data is lost or inaccessible, this is no small matter.

Software-defined storage: There are many open source distributed storage software such as ceph and swift, which can be built by themselves. The most stable device is the combination of software and hardware (but in essence it is the storage brand binding and cross-brand incompatibility), and the most pitfall is the pure software version (for example, VSAN, not only the purchased hardware must be in the compatibility list of VSAN, but also the driver and version need to be matched to the microcode level, otherwise compatibility problems may occur, which cannot be solved by opening a chassis). Of course, the open source version with the best scalability but the worst usability is. (Basically, command line operation is needed, and a very capable operation and maintenance team is needed, otherwise there will be no external support in case of failure. )