[Virtualization practice] 5 IOPS of storage design

Author: Fan Jun Frank Fan) Sina Weibo: @ frankfan7: frankfan7

IOPS is the most commonly used Performance Measurement Parameter in storage design. But isn't IOPS enough to help us make design decisions? There are many misunderstandings in this regard. In most cases, it is meaningless to consider IOPS alone. This article uses a case to show you how to fully consider all aspects that affect storage performance.

There are too many factors that affect the storage performance, and the features and processing methods of different storage products are also different. For the sake of simplicity, this article will not discuss Write/Read ratio, Sequential/random, RAID Penalty, Cache, Dedupe, auto-Tiering, Partition Alignment and other factors.

Metrics:

Throughput

The amount of data transmitted per unit time. It is usually measured in KBPS or MBPS.

Latency (Response Time)

The time required to complete an IO request. It is often measured by milliseconds.

IOPS

Input-Output Per Second. Measure the number of read and write operations that can be completed within one second.

Relationship between Latency and IOPS

IOPS = 1000/(Seek Latency + Rotational Latency)

The above formula is only used to calculate a single Disk. Of course, this does not reflect the real environment. But it helps us understand the relationship between the two. The data of disks is very old, which only indicates that the IOPS supported by different types of disks is different.

RPM	IOPS
SSD	6000
15 K	175
10 K	125
7200	75
5400	50

The above is the most simplified case, and the Write/Read ratio, RAID Penalty, Cache and other factors are not taken into account.

So when using a commercial storage array, can we say that the higher the IOPS, the lower the Latency? You cannot simply draw this conclusion.

IOPS and Latency are not a simple linear relationship. When IOPS is smaller than a value such as 3000, the higher the IOPS, the lower the Latency. Once this critical point is exceeded, for example, when the IOPS is at 4000, Latency may be three times the IOPS at 3000! This critical point depends on the array used.

When you see that an array supports Million IOPS Data, don't be too happy. Maybe the array handles so much IOPS when Latency will be up to 30 MS! If your application has high Latency requirements, such as the real-time Transaction of the credit card system, Latency greater than 4 ms is enough to make the Transaction fail.

Block Size, Throughput and IOPS

A common concern for virtualization architects is how much bandwidth the front-end interface card hba and iSCSINIC need? Can background storage handle the expected IOPS?

Suppose there is a bucket, and the volume of the bucket is certain. It can accommodate many small balls, but it can accommodate only a few large balls.

Here the volume of the ball refers to an operational BlockSize.

650) this. width = 650; "src =" http://www.bkjia.com/uploads/allimg/131228/01254L3K-0.png "title =" j1 (Mobile).png "/>

Block size:

The Block size is determined by the specific application. The Block Size of the same application varies under different circumstances. For example, when Oracle processes Online Transaction, the Block Size is 2 kb or 4 kb, while the BlockSize in demo-support system is 8 KB, 16 kb or even 16 kb. EXCHANG2007 uses 8 KB, SQL uses 8 KB at least, and SAP uses 64 KB

The following formula describes the relationship between Block Size, Throughput, and IOPS.

ThroughputMB perSecond) * 1024 (Convert KB to MB)/Block Size (KB/IO) = IOPS

Scenario 1

Taking a backup application as an example, for continuous write operations, the Block Size is 64 KB. If the 16 Gb fc hba is used, the Throughput is 1700 MBps. Apply this formula to obtain:

1700 MBps * 1024/64 KB = IOPS

Compared with 8 Gb FC, 16 Gb FC has a larger Throughput value, which means higher IOPS can be transmitted. Note that this only takes into account the frontend. Can backend storage handle so many IOPS? You need to consult your storage provider. While ensuring that this IOPS is supported, Latency must also meet your application requirements.

Scenario 2

Just now, we calculated the IOPS transmitted from the front end to the backend storage based on a certain number of Throughput conditions. We can also apply that formula in turn.

Assume that a virtual desktop environment using iSCSI shared storage has 20 IOPS for each win7 Virtual Machine and 200 IOPS for each. Therefore, the storage requirement is 20*200 = 4000 IOPS. Generally, the I/O size of a window is 4 kb. How fast does the iSCSI NIC take?

X * 1024/4 KB = 4000

X = 15.625 MBps or 0.12 Gbps.

This means that 1 Gbps iSCSI can fully meet the above requirements.

The formula is applied in either of the preceding two scenarios.

Refer:

StorageBasics-Part IX: Alternate IOPS Formula

Understanding how storage design has a big impact onyour VDI (updated September 2011)

Http://www.yellow-bricks.com/2009/12/23/iops/

UnderstandingIOPS

Moreon Performance Metrics: The Relationship Between IOPS and Latency

How many IOPS can a HDD, HHDD or SSD do with VMware?

This article from "sit up and watch the cloud" blog, please be sure to keep this source http://frankfan.blog.51cto.com/6402282/1214831