On February 19, 2014, NetApp launched the new FAS8000 series scale-out enterprise storage systems and virtualization software for the unbound Cloud Era.
How well the FAS8000 storage systems will handle the Exchange 2013 workload? In the months prior to the launch, we did some Jetstress testing on a 6-node FAS8060 cluster running Clustered Data ONTAP 8.2.1.
Figure 1 shows the high-level topology of the 200,000-mailbox in a virtualized environment created by using Windows Server 2012 Hyper-V. At the physical level, the testbed is composed of 6 Fujitsu RX300S7 servers connected to a 6-node FAS8060 storage cluster via a pair of Brocade 6510 Fibre Channel (FC) switches and dual redundant 8Gbps FC paths. Each pair of FAS8060 nodes and the Fujitsu servers form a building block to host one database availability group (DAG), which becomes the unit of scale. Each DAG has 16 virtualized mailbox servers (8 active and 8 passive). In total, 48 virtualized Exchange 2013 mailbox servers are organized in 3 DAGs.
Figure 1. High-level topology and DAG architecture of the 200,000-mailbox virtual environment.
The 200,000-mailbox virtualized environment consists of the following:
- 3 DAGs
- 24 active mailbox servers
- 1,390 mailboxes per database
- 8,340 mailboxes per server (6 databases per server)
- A total of 144 active databases for 200,160 active mailboxes
- A total of 144 passive databases, hosted by 24 additional servers; the passive databases are on separate aggregates and on the separate but identical storage controllers
- 2 copies of databases
The Jetstress 2013 targeted customer profile is as follows:
- 200,160 mailboxes
- 48 Exchange servers (24 tested)
- 0.05 IOPS (0.06 IOPS tested for additional 20% headroom)
- 1.0GB per mailbox
- Maintaining ample controller headroom available for storage infrastructure activities such as data replication and disk reconstructs
- 24/7 background database maintenance
- Mailbox Resiliency (2-copy)
Figure 2 shows the Jetstress performance test results on all 24 virtual Exchange 2013 mailbox servers.
Figure 2. Jetstress 2013 performance test results.
The achieved IOPS exceeded the target IOPS substantially, by 36% on average. The average database read latency was 10.66ms, below the required 20ms threshold. The average database write latency was 1.69ms, far below the required 20ms threshold.
Figure 3 highlights the scale-out capability of this configuration. It shows that by increasing the number of nodes inside the storage cluster from 2, to 4 and 6, the number of mailboxes supported also increased linearly, from 66,720, to 133,440 and 200,160.
Figure 3. The scaling of Exchange 2013 mailboxes.
For more details, please read TR-4268.
Thanks for reading.