Application of NFS protocol in storage

In many cases, we can use another method to solve problems or difficulties. Here we will explain the NFS protocol design of VMware's dynamic management platform. Most VMware-based vendors provide block-based fiber channel or iSCSI storage for their ESX servers, but some people have begun to propose Network File System NFS) it may be a more suitable choice. for example, when the NFS protocol designed a VMware-based dynamic service management host platform two years ago, IT commercial service provider T-Systems chose NFS.

Dr. Gregory Smith, Director of Dynamic Services, said, "We really think network NFS is the best choice." SAN has its advantages in throughput, but in terms of flexibility and management, NFS is much better ".

As an industrial standard shared file system, NFS volumes are easy to install and uninstall and support multiple clients. IT administrators also know about them. smith further mentioned the security advantages of NFS. because NFS runs over Ethernet, T-Systems uses virtual LAN VLANs to isolate Virtual Machine VMS one by one. "VMS only see their own things. you can place the file system on the LAN without considering the SAN, switch, and other things."

On the contrary, the VMS in the SAN's VMFS cluster file system currently do not support separating VM storage. If you want to use VMotion to migrate virtual machines through ESX hosts, it is more difficult. although ESX Server 3.5 that supports N_PORT ID virtualization will change this situation, NPIV is only supported by bare devices, while VMFS-VMware's proprietary file system does not.

T-Systems provides application services for approximately 70 servers and 80,000 end users. The system consists of NetApp network attached storage based on VMware's dynamic service platform. the company uses "a large number of [NetApp] OnTap software packages" Smith said, including using the remote image software Snappmirror to back up and using SnapVault to create a 30-day archive, metroCluster allows you to create a dynamic Computer Cluster between Distributed Remote Data Centers. It supports failover and copies data from the master site to the remote site to ensure that the data is completely new.

Is performance at the cost controversial?

But what is its performance? "When you tell someone that you plan to run VMware on the NFS protocol, people will think you are talking nonsense," said Scott Lowe, virtualization consultant for financial solutions and enterprise software company ePlus. this is because, from the perspective of swallowed up, NFS on Gigabit Ethernet cannot compete with 4G optical fiber. "If you only focus on throughput, the FC architecture beat NFS," Lowe said.

But Lowe and his partners pointed out an interesting paradox: the more VMS in an environment, the less obvious the FC throughput advantage. in fact, the actual performance of NFS in a large environment may be better than that of FC, Lowe said.

"When you start loading VMS in data storage, you get stuck in a SCSI congestion issue, as a result, preventing all open hosts from accessing LUNlogical unit number) "," the more hosts, the lower the throughput The FC can achieve. therefore, you often cannot reach the theoretical throughput height ".

However, not everyone agrees with this idea. "It sounds a bit too FUD-Fear), Uncerntainty guess) and Doubt suspect)", said Greg Schulz, founder and analyst of StorageIO Group. "Of course, if you build a system for debate, you may encounter some problems. however, if you follow the best operation method of the manufacturer, there should be no problem ". schulz said that most fc san disk arrays recognized by VMware will publish the "Operation Guide", detailing the system speed and data volume, and the recommended maximum number of hosts, maximum number of VMS per port on the array.

In addition, the debate on disks does not exist in SAN, Schulz said. the same thing can happen on NAS devices, unless you are discussing the number of threads or login records rather than the initiator and target ends of SCSI.

James Price, president of Fairway Consulting Group, is very clear against the performance advantages of NAS. "NAS has its place," he said. "But I don't think you can find any well-trained storage engineer who will tell you that NAS can be used as the primary storage platform ".

First, the debate on disks mentioned by Lowe may be a defect in the previous version of ESX, but in the new version of VMFS integrated with virtualization architecture 3, this problem is not so serious. now, Price explains that no longer all hosts in the same cluster share a log through disk access. Now it is equivalent that each host has its own dedicated log, eliminates communication congestion.

Second, Price is worried about adding a layer of Protocol due to a third-party file system NFS), leading to system complexity. "the real threat to NAS is block-level input/output I/O for other file systems," he said. "It's not visible from ESX ". in addition, you cannot perform many operations when using NFS to store data, such as out-of-band operations such as backup, cluster, SAN import, and bare metal ".

However, Lowe of ePlus said more users are following the NFS protocol path, including "a large number of users starting to migrate data from their top-level FC storage to NFS", he said. not to mention performance issues. Part of NFS's charm stems from the special features of the vendor's products, such as NetApp. "For them, backing up all the data is a huge challenge and cannot be solved. they have seen VMware's unified backup, but are not very satisfied. "" Lowe said. now they have decided to use NetApp's NFS volume, through which, "they can get data from the customers who have built it, and the backup is now much easier."