The PURE Joy of Flash
Until recently, the SSDs on the aspectra Shared ESX Farm were housed in the same storage as the HDD. The performance advantages of the SSD could therefore only be partially exploited. Thanks to the purchase of an all-flash array from Pure Storage, the IOPS could be significantly improved.
Storage Connectivity
Very high availability and data security are of paramount importance. The storage consists of an active/active cluster with two locally separated arrays, which are redundantly connected to the ESX farm via a Fibre Channel SAN. In this context, redundant means that each array has two controllers, which in turn are connected to two SAN fabric (switches) via two FC connections each. This connection method is known as FC Uniform Storage Connectivity.
Replication Connectivity
The synchronization of the data between the two arrays is done over the replication link. For this purpose, each controller hangs redundantly on an independent synchronization network via Ethernet.
Pod
Pure has introduced so-called "Pods" to simplify storage management. A pod is a management object that spans both arrays. It contains the actual volumes as well as the protection groups with the snapshot settings (interval and retention time). It is also possible to define a preferred array so that the VMs address their volumes on the nearest array if possible.
Each pod can be assigned a size that goes even beyond the physical capacity of all discs combined. Pure Storage Management manages the space occupied by the VMs and their snapshots, as well as the RAID configuration. A data reduction with different, combined methods increases the usable capacity. The data reduction factor depends on the application. In our setup in the server virtualization environment, a factor of 2.6 was achieved.
Mediator
The failover mediator is a third-party system that must be placed in a separate environment. Pure offers the Pure Cloud Mediator as a service. It is also possible to install a mediator as an on-site appliance. The mediator monitors the arrays and decides which of the two arrays continues the service if the synchronization fails. This prevents a "split brain" situation.
Failover
The aspectra storage team has tested the various error cases in a laboratory environment. Failures of single controllers or an entire storage array as well as interruptions of FC connections or the synchronization network did not lead to any problems. The virtual systems continued to run without any interruption or data loss.
Only the simulation of a multiple failure was able to put the virtual servers into freeze mode. First the synchronization was disconnected and after a certain time an array (B) was switched off. The VMs were frozen and had to be restarted on array (A) with the data from the time before the synchronization failed.
Monitoring
Hardware failures are detected by the mediator and reported to Pure Support. They will automatically replace the hardware. The Pure-Support also reports to the hosting provider, if the replication link fails, for example. In the Pure Web Client there is a menu item for analysis functions. We have also tapped into the REST interface to monitor the storage in our SNMP-based monitoring environment. This API also uses the available templates for Zabbix, which we use in parallel.
Read the latest on Flash-Array in the Pure Storage Blog.