This article describes how to use and build a highly-available storage that ensures business continuity even in the event of total storage server failure.
It will also explain you the main differences and advantages/disadvantages of euroNAS TCP/IP based HA Cluster and Fibre Channel SAN Cluster that uses fibre channel protocol.
Until recently, building a highly available, performance fibre channel storage that provides high availability was very expensive. euroNAS changes this by providing a powerful, scalable high-availability solution that is both affordable and easy to deploy. It brings to small and midsize companies the enterprise functionality, usually affordable only to large companies
What makes euroNAS Cluster different from other solutions?
On the market there are many SAN solutions providing high speeds and redundancy. They will protect you from disk failures or failure of one RAID controller.
euroNAS goes a step further by protecting you from a total failure of one of the units and also enabling you to simultaneously have the same data on 2 different places.
This will protect your data in case of some catastrophic events like fire, earthquake or theft and provide you business continuity at the same time.
Also not many manufacturers on the market will offer you both types of server mirror (TCP/IP and Fibre Channel) enabling you to use the same solution provider depending on your budget and performance requirements.
Difference between Ethernet (iSCSI & NFS) and Fibre Channel
iSCSI and NFS are protocols that use TCP for transport and enable you to use the existing TCP/IP network infrastructure. iSCSI has been developed as cost efficient alternative to fibre channel. It is built on an underlying TCP/IP protocol and is usually implemented only as a software initiator that incurs significant processing overhead. Similar to Fibre Channel, iSCSI will present its targets to iSCSI initiators directly as block devices. NFS will present the device at the file system level.
Greatest benefit of iSCSI is the cost of the solution. In comparison to FC, iSCSI requires less expensive hardware and since it is based on Ethernet, more IT Administrators are familiar with the technology. However, Ethernet is not designed for transferring block data in a networked storage environment.
One of the greatest problems of the Ethernet is its way of handling data collisions – when more than one computer tries to transmit data simultaneously. Under heavy load condition, too many packet collisions will greatly reduce the whole network efficiency due to retransmissions. When an iSCSI or NFS path is overloaded, the TCP/IP protocol drops packets and requires them to be resent. FC has a built-in pause mechanism when overload occurs. On iSCSI and NFS this can lead to oversubscribed network paths and the performance degrades more and more because the dropped packets need to be resent.
On iSCSI and NFS this problem can be reduced by lowering the traffic, bonding the network ports or using switches with larger port buffer.
Fibre channel with its asynchronous protocol design ensures that even under heaviest load, collisions are handled efficiently and maintains maximum throughput.
While fibre channel constantly performs very close to its max speed ( 8 Gb = 800 MB/s or 16 GB = 1600 MB/s), collision management on Ethernet makes it very difficult to achieve same consistent performance.
The bottleneck of euroNAS Cluster is the disk I/O and the speed of the FC/Ethernet controller. We recommend the use of SAS or SSD disks and hardware RAID controller.
euroNAS Cluster is hardware independent however we still recommend using identical hardware. Due to server replication in real time, the performance will depend on the weakest link. Using slower disks on one server will slow the server with faster disks as well.
Both servers are proven and are working reliably in most demanding VMWare environments. However, HA Cluster and SAN cluster differentiate in failover concept.
HA Cluster will provide its resources using the virtual IP connection. If failover event occurs, it will move this virtual IP Address to the working node. VMWare ESX is a cluster aware OS and will wait for the failover to finish. VMs can experience small freeze during failover but will continue to work normally after that.
Most Servers/Clients supports ALUA/MPIO multipathing, multipathing allows you to define more than one physical path that transfers the data. In case of failure, ESX will automatically switch to another working physical path.
SAN Cluster will communicate to the Server using Asymmetric Logical Unit Access (ALUA) – it will notify the server which node it should use and which one should be in standby. If failover event occurs, remaining port will automatically register itself as active, VMware will automatically continue to use this port. VMs can experience small freeze during failover but will continue to work normally after that. Usually the freeze is much shorter than on HA Cluster.