Be sure to read through the previous post for a quick introduction of IOPS.
Suppose that you’re copying a 10GB file from one disk to another. Let us assume that the only bottleneck is the disks’ own IOPS, which in case is 120. We are also only focusing on the disk where the data is copied to and assume that the other disk has no IOPS limitation. Because we are writing about 10GB, the majority of the IOPS are going to be writes and not reads.
We will also assume that the disk has a block size of 64KB. In other words we are copying a maxium of 64KB of data in one operation (120 operations per second). That gives us a sustained throughput of about 7MB/s (120 * 64 * 1024). For a 10GB file, it would take about 1462 (10GB/7MB) seconds to copy the 10GB file.
UNIXy Uncategorized bandwidth, example, IOPS, throughput
IOPS stands for Input and Output Per Second. It is essentially a value that describes the raw capacity of a data storage system. It sets an expectation for performance. Some storage systems are said to be capable 120 IOPS and others 340 IOPS or much more. The number of IOPS, for most disk systems, represents a count of a mix of data reads (Output) and writes (Input) per second. By that number, one can judge how fast and responsive a disk system can be. The numbers can be obtained by performing sequential or random operations on the disk system. The best representation is random operations since it reflects a real-life usage pattern.
Disk system manufacturers obtain IOPS numbers via benchmarking. There are tools that measure disk throughput and capacity by performing several read and write tests. For example, Iometer is a tool that can measure disk system performance and produce the IOPS value for such system. One needs to be aware of the fact that some manufacturers provide the IOPS based on cached read and write tests. The latter is a best case scenario test and does not reflect real-life usage.
So what happens when a system reaches the IOPS threshold number? I did use the word “reach” because in theory the disk system will almost never exceed the advertised IOPS. Back to the question, when the system reaches the IOPS threshold, it starts queuing requests. A busy queue generally indicates a bottleneck.
How often have you found yourself with a bottlenecked 2TB disk choking after barely putting a few files? Disk size capacity has no bearing on its IOPS capacity. A 2TB disk could have the same IOPS capacity as a 250GB disk! So when does one need to upgrade the disk system or the configuration at least? We shall post a new entry on this blog with the instructions.
That’s all folks!
UNIXy Crash Course explained, IOPS
