CASE STUDY: PUBLIC UTILITY CUSTOMER

Load DynamiX enables public utility company to select optimal storage array

Storage Array Bake-Off

Storage and storage network performance is critical to any industry which spends a significant budget on data storage. Public utility companies must keep up with bandwidth-demanding applications, and a user community that increasingly expects better service, in the face of intense scrutiny of costs and business margins.

To keep pace, storage architects within the public utility industry must wring the best performance out of existing platforms, and make well informed decisions on new storage deployments. Storage performance testing and validation is becoming a key component to cost-effective and responsive storage system rollouts. The primary goals are to reduce the risk of organizational interruption in the form of outages and performance issues, and to ensure the right product and amount of product is acquired. More specifically, the evaluation of storage infrastructure should include modeling the production application workload environment as accurately as possible. It should also enable the ability to pre-determine the performance limits of the storage system so that the next storage upgrade can be a carefully planned event instead of a disruptive fire drill.

The Customer Challenge

“We were quite pleased with the process and results of the storage array tests. Load DynamiX worked closely with us and the vendors to show their products fairly and independently.”

— Infrastructure Manager,
Public Utility Company

This public utility company manages the flow of electric power to over 10 million customers. As the independent system operator, it schedules power on an electric grid that connects more than 20,000 miles of transmission lines and 200 generation units. It is subject to oversight by a Public Utility Commission and the state legislature. With the growth of data and the aging of its existing infrastructure, the company was investigating the addition of a large fibre channel storage system. After shortlisting three leading vendors, whose products all met the functional criteria including capacity upgrades over the life of the arrays, the decision came down to one of price/performance. The vendors all offered Proof of Concept systems, and the company began looking for a valid capability that simulates storage workloads representative of their production application workloads. Without the luxury of an in-house staff of architects, the decision was made to outsource the performance evaluation project.

Load DynamiX Solution

After an online demonstration of the Load DynamiX capabilities and a conversation with an existing customer, the company decided to go with a Load DynamiX professional services engagement. A suite of test projects was co-created that (1) benchmarked performance under representative workload parameters, and (2) provided “what if” analysis that showed performance under 2X, 3X and 4X loads.

Storage Performance Comparison Process

The customer worked with the Load DynamiX Professional Services team to define the performance tests, deploy the Load DynamiX appliance, run the tests, and analyze the results to empower the customer to make intelligent deployment decisions. The comparison was done by measuring performance characteristics (IOPs, throughput and latency) corresponding to the workloads generated by the Load DynamiX appliance.

The company then provided historical I/O performance data collected over two months. Data included one minute averages of Reads and Writes per second, Read and Writes MBs/second and Read and Write hit percentages.

Tests were then set up as in figure 1 below, so that each of the four FC ports on the Load DynamiX appliance accesses LUs on the arrays, also called Systems under Test (SUT), through different paths. The configuration of the paths was specific to every vendor. Each SUT had 30 TB of capacity provisioned for testing: six LUNs, 4 TB each and 24 LUNs 256 GB each. Each of the first three ports on the appliance targeted two LUs, 4 TB each. The last port (Port3) targeted the smaller LUN.

Figure1-300x146

Figure 1: Diagram of the public utility customer’s test system.

 

Additional details of the customer test environment included:

  1. The custom workload consisted of two Load DynamiX scenarios, one for read and another for write operations assigned to each of the ports.
  2. A 4-step overload workload was used as a stress test to show results when the system was more heavily loaded. Throughput was stepped to 400, 800, 1200, and 1600 MB/s by incrementing IOPS.
  3. All arrays were preconditioned by writing data to 90% of the LUNs. There were 30 LUNs used on each of the arrays with the total capacity of 30TB.

 

Test Results

The Throughput over Time chart shows the performance difference between the 3 arrays in the most dramatic way (see figure 2 below):

  • System 3 performs nicely for levels 1 and 2 and maintains high level of throughput at level 3 while not being able to achieve level 4 (throughput tops out at level 3).
  • System 2 performs well on level 1 but not able to perform as high as level 2 (never achieves same throughput from 2 – 4X load).
  • System 1 does not show enough performance capacity to achieve level 1.

 

Figure2-300x161

Figure 2: Throughput over time for a 4-step overload test. Note that the throughput did not increase for any of the systems under test from 3.0 – 4.0 hours for the 4th overload test.

“With Load DynamiX, architects can identify performance limits for their unique workloads and determine the strengths and weaknesses of any networked storage array.”

— Howard Marks, Chief Scientist
DeepStorage.Net

Another view is the custom workload 12-hour test. In the example below, at double the load performance, systems 2 and 3 are almost identical (system 2 graph is hidden behind system 3). System 1 is not achieving the same level of throughput.

 

Figure3-300x171

Figure 3: Throughput over time during the 12 hour 2X load test.

 

Figure 4: Average Read Latency over time.

Figure 4: Average Read Latency over time.

 

Figure 5: Average Write Latency over time.

Figure 5: Average Write Latency over time.

Summary of Results

The results realized after Load DynamiX professional services performed comparative testing of three different vendor storage arrays at the customer’s site included:

  1. All three arrays showed similar performance for a production level workload in terms of throughput and write latency. Read Latency showed high values for System 1.
  2. At 2x and 3x production load, Systems 2 and 3 outperformed System 1 on all metrics and were similar in performance.
  3. At the 4x overload test, System 3 showed superior latency performance over Systems 1 and 2. System 2 showed better performance than System 1. None of the systems showed an increase in throughput over the 3X test.

Benefits of Load DynamiX Approach

This public utility customer realized the following benefits after using Load DynamiX:

  • Confirmation of which storage arrays would offer the best performance today, and critical insight into storage headroom when increasing data loads.
  • Storage performance assurance today and truly predictable performance that eliminates unnecessary under and over-provisioning in the future.
  • Confidence gained knowing that the selected storage system will support the workloads in their production data center.