[Virtualization practice] Disaster Tolerance Design 4 VPLEX

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

The emergence of storage devices such as VPLEX enables dual-active data centers to maximize the effective use of computing and storage resources.

In "Disaster Tolerance Design 3 Stretched Cluster", we introduced the concept of Stretched Cluster. emc vplex is one of the storage devices supported by the Stretched Cluster. This article describes how to integrate emc vplex with VMware Stretched Cluster. Of course, VPLEX can also be used independently.

IVPLEX Introduction

650) this. width = 650; "src =" http://www.bkjia.com/uploads/allimg/131228/002621N26-0.png "title =" vplex (Mobile).png "alt =" 141722111.png"/>

The requirements of the Stretched Cluster for storage are that in the case of any disaster, virtual machines can access data without affecting the performance. After the disaster recovery, the storage device should be able to automatically switch to the initial setting as much as possible to avoid service interruption due to excessive manual intervention.

Let's see if VPLEX can meet these needs.

Taking VPLEX Metro as an example, the device has three outstanding features:

Compatibility:VPLEX not only manages EMC storage devices, but also devices of many other vendors.

Distribution:VPLEX integrates underlying storage resources and creates an intermediate layer called Distributed Volume. VSphere Cluster does not need to know what storage devices are used at the underlying layer, but only needs to create a Data Store on Distributed Volume. Distributed Volume supports VMFS and RDM.

Intelligence:Assume that a vm on Site A writes data on Distributed Volume, VPLEX completes the write operation on the storage devices on Site A and Site B at the same time. Note that write operations are initiated by the local VPLEX device to the local storage, so there is no latency caused by Remote Write. Read operations are performed locally.

IIArchitecture Decisions

Suppose you have decided to use the Stretched Cluster. In the evaluation of VPLEX, you need to consider the following important Architecture demo-

Which one to choose?VPLEX Device

Type	Comments
Local VPLEX	Applicable only to a single data center
Metro VPLEX	Usually used for two data centers with a distance of less than 100KM, Synchronous up to 5 ms RTT
Geo VPLEX	Usually used for two data centers with a distance of less than 100KM, Asynchronous up to 50 ms RTT

Required?Witness, how to set Witness?

For example, if the FC or IP network used for communication between VPLEX devices is interrupted between data centers, since both data centers can initiate write operations locally, this is not a mess. How to determine which data center's storage device is the primary device preferred) and which data center's storage device is from the non-preferred device ). This is usually called "Split Brian ". As the name implies, two headquarters may appear, working separately, which may result in data inconsistency between the two locations.

To avoid the above situation, Witness VM plays the role of the referee to determine which Site is Preferred Site. Note that Witness VM is recommended to be deployed in different physical locations than Site1 and Site2. Services cannot be interrupted at the same time by a Site and Witness VM.

In some situations where the specific environment or risk is acceptable, Witness can also be disabled.

Non-uniform Host AccessOr uniform Host Access

Non-uniform Host Access indicates that the ESXi Host of SiteA can only Access the storage devices of SiteA. The same applies to SiteB. ESXi hosts cannot access remote storage devices.

Uniform Host Access means that the ESXi Host can simultaneously Access local and remote storage devices.

Iii. Possible impact of the above decisions

Failover Scenario

All components that may be interrupted must verify the consequences of the interruption. For example:

A single ESXi instance is down, multiple ESXi instances are down, the entire data center is down, a single ESXi network is interrupted, and a network is interrupted for storage synchronization between data centers. StoragePartition) two-layer network interruption between data centers DataCenter Partition), a single VPLEX device is interrupted, multiple VPLEX devices are interrupted, the storage array disk is damaged, the entire storage array is down, and the entire data center is interrupted. For details, see this article.

Metro Cluster FC WAN

Pay close attention to the traffic of FC or IP networks connected to VPlEX devices in two locations. When you create a VPLEX Metro Cluster for the first time, you need VPLEX Sync to synchronize data between two locations. Depending on the amount of data to be synchronized, network saturation may occur. If this link is shared with other applications except for VPLEX, it may affect other applications.

In this case, you need to plan the time period for VPLEX Sync in advance.

Refer:

Implementing vSphere Metro Storage Cluster (vMSC) using emc vplex (2007545)

Emc vplex 5.0 ARCHITECTURE GUIDE

Conditions for stretched hosts cluster support on emc vplex Metro

This article from "sit up and watch the cloud" blog, please be sure to keep this source http://frankfan.blog.51cto.com/6402282/1298689