Shulou(Shulou.com)06/01 Report--

Today, I will talk to you about how to take ScaleIO as an example to analyze the deployment of SERVER SAN. Many people may not know much about it. In order to make you understand better, the editor has summarized the following contents for you. I hope you can get something according to this article.

Actually, SERVER SAN is good, but deployment is troublesome. It involves operating system, network and other aspects. Especially in the planning of the network, the performance will be extremely poor, and even bring the application to death in the case of super-convergence deployment.

Today we intend to take EMC ScaleIO as an example to talk about the deployment of SERVER SAN. The following share is the general consideration of ScaleIO deployment.

Introduction to ScaleIO

In fact, when deploying any SERVER SAN, we first need to understand the basic framework of the SERVER SAN. Different vendors adopt different technologies, so they are deployed in different ways.

Although we have shared ScaleIO before, let's do a brief review.

ScaleIO software has three main components:

Meta Data Manager-MDM, the metadata management component, needs to be deployed as an active and standby cluster mode. In addition, one needs to be deployed for Tie Breaker to decide which process to take over the metadata management when the cluster is split.

ScaleIO Data Server-SDS, a data server, provides storage services.

ScaleIO Data Client-SDC, the data client, installed on the host side, and hosts that need to access ScaleIO storage all need this thing.

If you have learned some distributed file system architecture, it is easier to understand this, because generally for good scalability, there is a metadata layer. For example, Kunteng's StorNext, although a file system, also has these three components (when using a dedicated NAS client). Huawei's OceanStor 9000 is similar, but it is not implemented separately because it uses a standard NAS client and the metadata components are managed by the system itself.

Since ScaleIO requires a dedicated SDC client, we can understand that this is not a general SERVER SAN. Unlike Huawei's FusionStorage and other SERVER SAN, it does not provide one or more general ISCSI or FC host interfaces, but needs to install a proprietary client on the host. Although this proprietary connection is more troublesome to deploy, it is more convenient to coordinate with private protocols. It also avoids the trouble of dealing with the multipath switching of traditional IP SAN or FC SAN.

After understanding these three modules, we can better plan the network. The network is divided into management network (access MDM part), control network (inter-MDM interconnection part), reconfiguration network (inter-SDS interconnection part), data path network (SDC access SDS part). If there are conditions, the four networks are separated, the traffic between each other does not affect each other, good performance, reliability and security are also more guaranteed. Of course, there is no problem with the test environment being deployed together.

In ScaleIO, there is also the concept of a protection domain, a protection domain is a set of SDS that protect each other. A SDS can only belong to one protection domain. The protection domain can be divided into many storage pools. Generally, the media of a storage pool is the same. For example, the performance pool uses SSD, while the capacity pool uses HDD. Although the storage pool can also be configured with mixed media, because ScaleIO does not have the concept of automatic tiering, the performance of this storage pool can only be calculated according to the slowest media, which is too wasteful, so such deployment is not recommended.

There is also a logical term called fault-tolerant set. This is also divided by users themselves, generally according to the power supply line, it is common for a cabinet to do a fault-tolerant set. As we all know, ScaleIO uses two copies of data protection. After dividing the fault-tolerant set, you can ensure that the two pieces of data are saved in different fault-tolerant sets, so that if the power is lost in a cabinet, the data will not be lost.

MDM,SDS,SDC is logical, so when physically deployed, it may be on different physical nodes or on the same node.

General requirements of ScaleIO for Network and OS

The network requirements of each node in ScaleIO need to be paid attention to:

One of the following:

1 gigabit or 10 gigabit (recommended) network

IP-over-InfiniBand network

Dual-port network interface cards (recommended)

Ensure the following:

There is network connectivity between all components.

Network bandwidth and latency between all nodes is acceptable, according to application demands.

Ethernet switch supports the bandwidth between network nodes.

MTU settings are consistent across all servers and switches. For jumbo frame support, set the MTU for servers, switches, and vSwitches to 9000.

The following ports are not used by any other application, and are open in the local firewall of the server:

-MDM: 6611 and 9011

-SDS: 7072

-Tie-Breaker: 9011

-ScaleIO Gateway (includes REST Gateway, Installation Manager, and SNMP trap sender): 80 and 443

-Light Installation Agent (LIA): 9099

The following port is open in the local firewall of the server:

-SNMP traps: 162

Although ScaleIO supports Ethernet and Infiniband, IB is generally rarely used. Generally, a node needs two 10GE ports. When planning for network security, we should pay special attention to the IP port number of each application, and ensure that its messages can be passed when the firewall is configured.

ScaleIO's requirements for the operating system should also be checked in advance:

One of the following (for a complete list, see the EMC Support Matrix):

Linux: CentOS 6.07.0,6.07.0 Red Hat 6.07.0,11 SP2 and SP3, or SUSE 12

Packages required for all components:

-numactl

-libaio

Packages required for MDM components:

-mutt (for Call-Home)

-bash-completion (for scli completion)

-Latest version of Python 2.x

-When installing the MDM component on Linux CentOS 6 or RHEL 6 hosts, set the shared memory parameter in the / etc/sysctl.conf file to at least the following value:

Kernel.shmmax=209715200. To use this value, type the sysctl-p command.

Requirements for running the GUI:

-Java 1.6, or higher

-Screen resolution: 1366 x 768 minimum

Windows: 2008 R2, 2012, or 2012 R2

Requirements for running the GUI:

-Java 1.7, or higher

-Screen resolution: 1366 x 768 minimum

Packages required for MDM components:

-Install the EMC-provided PythonModulesInstall.exe on all MDM nodes. Download the file from the EMC Online Support site (search for ScaleIO Python Installation Modules) on https://support.emc.com.

To install SDC on 2008 R2, ensure that Microsoft Security Update KB3033929 is installed.

Hypervisors:

-VMware ESXi OS: 5.5 or 6.0, managed by vCenter 5.5 or 6.0 only

-Hyper-V

-XenServer 6.1

-RedHat KVM

We see that ScaleIO supports Linux and Windows operating systems, as well as VMware, Microsoft, XenServer and Redhat's Hypervisor, which is quite comprehensive. Among the commonly used OS, only Ubutun does not support it.

Super-converged deployment of ScaleIO

ScaleIO supports both layer 2 deployment (SAN.NEXT) and superconverged deployment (INFRASTRUCTURE.NEXT).

In super-converged deployments, the most common ones are VMware,XEN and OpenStack.

1. Combine with VMware ESX.

Although ScaleIO is not integrated into the vSphere kernel like VMWARE VSAN, SDC can still be deployed in Hypervisor, but SDS needs to run in a separate VM (SVM). VM can use storage through VMFS or RDM.

2. Combine with Xen VM architecture.

ScaleIO can deploy both SDC and SDS in DOM 0, but it is recommended that you increase the memory of DOM 0.

3. Deploy in conjunction with OpenStack.

ScaleIO provides Cinder interface and Nova interface, which can be deployed in OpenStack environment. Cinder interface is easy for everyone to understand, it provides block services, because ScaleIO is a SAN after all. But why need to provide a Nova (providing computing) driver? General storage does not need this driver ah? This is mainly because ScaleIO needs its own SDC client, so the SDC must be controlled through the computing interface Nova, so that the corresponding instance can access the corresponding volume.

After reading the above, do you have any further understanding of how to analyze the deployment of SERVER SAN using ScaleIO as an example? If you want to know more knowledge or related content, please follow the industry information channel, thank you for your support.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.