The best of all worlds: Dell EMC Ready Bundle for Oracle (Part II)

In my last blog I talked about the small and medium configurations. In this blog, I am going to talk about the large configuration of Ready Bundle for Oracle (RBO) .Part III of this blog series will talk about the different backup options available in this product. To back up the small configuration of RBO (called commercial backups), we use Data Domain 6300 while Data Domain 9300 (DD9300) is used for backing up large configuration (called enterprise backups). Let’s talk about the large configurations in this blog here. It is very exciting to talk about the large configuration which mainly caters to the enterprise world which wants to use the oracle environment in a bigger scale with higher IOPS and bandwidth along with lesser CPU utilization and latency. The large configuration of Ready Bundle for Oracle hosts 50 mixed-workload databases on a vSphere cluster with two ESXi 6 hosts on PowerEdge R940 servers. A VMAX 250F array with two V-Brick blocks, 2 x 1 TB mirrored cache and 105 TBe SSDs is used as the storage array for the VM OS and Oracle RAC databases. Figure 1 depicts the large-configuration architecture.

Fig. 1: Architecture of large configuration

The large configuration uses PowerEdge R940 servers, which are designed for larger workloads, to maximize enterprise application performance. The PowerEdge R940 servers are configured with 112 cores—88 more physical cores than the PowerEdge R740 servers in the small configuration and 40 more physical cores than the PowerEdge R940 servers in the medium configuration. The PowerEdge R940 server is configured with 3,072 GB—2,688 GB more than the R740 server in the small configuration and 1,920 GB more than the R940 server in the medium configuration. In Figure 2 we have performed a comparative analysis of the configurations for the small, medium (discussed in my previous blog) and large configurations of RBO.

Fig. 2: Comparative Analysis of small, medium and large configurations of Ready Bundle for Oracle

The comparative analysis will help the customers to select the appropriate configuration based on their requirements. There is also a tool available called VMAX sizer (available with the local support rep) which helps the customers to ascertain the appropriate configuration for their database/data center infrastructure. Now, I would like to discuss about the spectacular performances that were achieved during the stress-testing of large configuration for RBO. Let me provide some background for in-house database stress-testing which was performed by DellEMC engineers. The engineers ran 10 OLTP production RAC databases, 30 OLTP development RAC databases, and 10 OLAP RAC databases in parallel for a total mixed workload of 50 RAC databases. SLOB was used to create a 60/40 read/write OLTP workload. Swingbench Sales History benchmark was used for the OLAP workload generation. By adding the OLAP databases, they tested the VMAX 250F array along with large sequential reads to the storage workload. This load is reflected by the database size of 2 TB and the db_32k_cache_size and db_file_multiblock_read_count settings that enable larger database I/O to improve large sequential read performance. The entire architecture and database configuration of this use case has been depicted in the figure 3 and 4.

Fig 3: Architecture of the use case for large configuration

Fig 4 : Oracle RAC configuration for the use case in large configuration of RBO

The large configuration included two PowerEdge R940 servers and a VMAX 250F array. Here is a review of the results for this use case, in which the workload of 10 production OLTP databases, 30 development OLTP databases, and 10 OLAP databases ran in parallel:

• 100 vCPUs and 2,640 GB vMem per R940 computer server were used to generate a workload of over 189,968 IOPS on the VMAX array as is depicted in figure 5.
• Due to the processing power of the PowerEdge servers, the CPU utilization was only 20 percent, leaving room for more databases or for failover of VMs from one ESXi server to another.
• The VMAX array with inline deduplication and compression saved 5X the flash space, using only 8,280 GB of capacity for 30 development databases.
• Database IOPS was also excellent. The 189,000 IOPS of workload (highlighted in figure 5) were serviced at sub-millisecond latency.
• The 10 OLAP databases generated a total of 3.88 GB/s of throughput.

Fig 5. Aggregate IOPS numbers for the use case of large configuration

Figure 6 below illustrates the complete dashboard of the configuration and performance numbers (including the results that were discussed above) that got generated after the stress-testing for this use case in large configuration of RBO.

Fig 6. Stress-testing Dashboard of the use case for large configuration

The test results show that this configuration of RBO offer twice the number of supported IOPS compared to the small and medium configurations. The salient feature of this large configuration is that it supported just over 189,968 IOPS below sub-millisecond latencies for all OLTP databases. It supported the total bandwidth of 3.88 GB/sec using two PowerEdge R940 servers and a VMAX 250F with more flash drives. Under this database workload, a mere 20 percent of the CPU capacity was utilized, while the VMAX array accelerated 100 percent on most of the writes. Like the small and medium configurations, storage efficiency continued with 5X space savings across the 30 development OLTP databases due to the inline compression of VMAX 250F. In addition, the large configuration still had plenty of unused resources to support even greater workloads. In the next blog (Part III), I would like to discuss about a new Oracle solution on XtremIO X2 Array.