Stripe depth Queue Depth NCQ IOPS

http://blog.csdn.net/striping/article/details/17449653

IOPS is the I/O per second, which is the number of read-write (I/O) operations per second, used in databases and other applications to measure random access performance.

The concept of concurrent IO: Concurrent IO means that multiple IO can be processed at the same time, such as IO1 can access a-disk, IO2 can access the B-drive at the same time. The antonym of concurrent IO is sequential io.

stripe Depth : raid5 128KB stripe, 128KB stripe = number of disks multiplied by the segment size of this stripe on each disk, that is, a strip of multiple disks arranged horizontally into one piece, the hard disk itself is equivalent to a vertical bar, and the bar and the vertical strip formed after the small lattice is segment, also called strip depth, stripe depth. For example, the 8-disk RAID5 system, where a piece is used to store the parity,128kb stripe divided by 8 equals 16KB, which means that the segment= stripe depth =16kb= the space used to contribute a stripe on each disk.

Queue Depth : describes the maximum I/O values that are activated at the same time per disk . For example, in your configuration, each LUN has 4 disks, and if your queue depth is defined as 16, then the maximum number of I/Os that the entire system has at the same time is 64 (4 disks * 16 maximum I/O values per disk);

The queue depth is the number of commands in the command queue that are hosted on a single device.

The queue depth is the larger the number represents, the more tasks you want to process, the more concurrent

HBA Queue Depth refers to the amount of storage I/O that is sent to the data storage infrastructure for processing. When installing an HBA, most storage administrators only use the HBA card's default settings, and the default HBA queue depth setting is generally too high. This causes congestion on the storage ports, which can lead to application performance issues. If the queue depth is set too low, the port and SAN infrastructure are not efficiently used. When the storage system does not load enough I/O waiting to be processed, the opportunity to use its cache is not available. However, if almost all of the data in the cache expires before it is accessed, most data access will come from the hard disk. Most HBAs set the default queue depth between 32 and 256, but the best range is actually close to 2 to 8. Most start-up programs can report the number of requests that are not processed in their queues at any moment, which allows you to strike a balance between deep and shallow queue depths.

NCQ Technology It is a way to optimize the workload execution sequence within the hard disk by reordering the commands in the internal queue for intelligent data management and improving the performance constraints of the hard disk due to mechanical components. NCQ technology is an important part of the Sataⅱ specification and is the only technology related to the performance of the Sataⅱ specification.

The NCQ Technical SATA specification supports many new features, one of which is the NCQ (Native command queuing full-speed commands queuing) technology. First, let's look at how hard drives read and write information. The hard disk saves information by writing information to a specific location on the disk track

The process for accessing information on disk is as follows:

Locate the target disk (platter) that stores the data and access the disk.

Look for the track that stores the data on the disk and access the track.

Find the target cluster (cluster) that stores the data on the track, and access the cluster.

Find the target sector (sector) of the book data on the cluster, and access the sector.

Find the target data and read the data.

Through the above steps, the hard disk can obtain the required data information. The same is true of the steps for writing data to a hard disk, except that the read operation becomes a write operation. In most cases, the data is stored in the hard disk instead of sequentially, and it may even be possible to allocate a file to a different platter. For a hard disk that does not support NCQ, a large number of data read and write needs to repeat the above steps, and for different locations of data access, the head needs more operations, reducing the access efficiency. The hard drive that supports the NCQ technology is re-arranged according to the distance of the address they are visiting, so that the implementation of the mechanical action of the hard disk can be carried out intelligently in-house management, which greatly improves the efficiency of the whole workflow: Take out the commands in the queue and reorder them to efficiently get and send the data requested by the host. , while the hard disk executes a command, a new command can be added to the queue to wait for the job to execute. Obviously, the command arrangement reduces the time the head arm moves back and forth, making the data read more efficient.

As shown, if the new command happens to be the most mechanically efficient processing, then it is the next command to be processed in the queue. For example: For example, to release a set of data transfer instructions to the hard disk, because the data on the disk distribution bit different, the head may read 260 sectors, then read 7660 sectors, and then read 261 sectors ... If we optimize the order, we can read 260 sectors first, then read 261 sectors, then read 7660 sectors. Obviously, the command arrangement reduces the time the head arm moves back and forth, making the data read more efficient. In addition to considering the linear position of the target data, the effective sorting algorithm also considers its angular position and optimizes the linear position and angle position to minimize the service time of the bus, which is also called "reordering of commands based on seek and rotation optimization".

Stripe depth Queue Depth NCQ IOPS