Hardware Considerations and Performance Recommendations
The hardware used is as important as the software. Systems like the JMR SHARE is well optimized to run with euroNAS HA Cluster OS, which is a very robust Linux-based operating system that has proven extremely reliable. Using separate mirrored system disks, even that fault possibility is covered. In the JMR SHARE, a 3.5 GHz multicore CPU and 16GB high speed RAM, along with an internal 6Gb hardware SAS RAID controller (usually set to RAID 50 or RAID 60) allows each disk array to perform at about 1,800 MB/s internal transfer rates.
This is good news for prospective users, but the corporate network can be the choking point for any shared storage system. While GbE is common and multiple network ports are provided by the SHARE server, 10GbE obviously allows greater bandwidth and faster operations. The SHARE server provides dual 10GbE ports in addition to GbE, and these serve to bond not only the corporate network but also the two servers to each other (the private replication network).
A JMR SHARE HA Cluster system consists of two identical servers with HA Cluster installed on mirrored OS drives, the RAID storage drives are initialized and ready for your data, and the provided servers are tested in cluster configuration.
Limitations
There are some limitations to all systems, and the HA Cluster has a few:
• Replication network failure: Although the corporate network may remain operational, if the replication network which provides the synchronization “heartbeat” between the nodes fails, the cluster will not perform properly. To prevent this occurrence, we always recommend network card teaming (port failover), and provide for this with the SHARE system.
• Powering down master node during synchronization process: During the sync process, if the node with more
current data is powered off, server replication service will cease to prevent data corruption. As such, only the server (node) with less current data may be powered off; however, if both servers are fully in sync, either of the servers may be powered off or re-booted.
• Port failover (active backup) must be used for configurations without a switch: If the replication network is set to switchless configuration, the only network bond that may be used is port-failover (active backup); it is the only one that does not require any switch.
Various system configuration diagrams follow
(see pictorial diagram)
54