PowerVault 715N Storage Server: Scalability and Performance
By Julia Courtney and Aaron Burns (August 2002)
Dell ran scalability and performance tests on the DellTM PowerVaultTM 715N server, an entry-level network attached storage (NAS) system. The scalability test observed how increasing the number of simultaneous users affected system resource utilization. The performance test compared three PowerVault 715N models and determined the optimal network configuration.
Low cost and ease of use are two primary reasons organizations buy low-end network attached storage (NAS) systems. These "headless" devices are simple to deploy into an existing workgroup, small office, or small business network and can be easily managed from any browser.
The DellTM PowerVaultTM 715N storage server is the entry-level member of the Dell family of NAS systems. It has a Microsoft® Windows® Powered operating system, Dell OpenManageTM integration, point-in-time capture software, dual 10/100BaseT Ethernet ports, and front-loading, easily serviced drives. It also includes a SCSI port for local, LAN-free tape backup. To better understand the capabilities of the PowerVault 715N, Dell conducted several scalability and performance tests.
Scalability of the PowerVault 715N
To size the PowerVault 715N storage server, Dell used Mercury Interactive® LoadRunner® version 7.02, a leading load-testing product. One of the software's most important features is its ability to emulate many clients on a single system. LoadRunner 7.02 supports several well-known protocols and major software packages used today. For testing, Dell used a combination of Microsoft PowerPoint® , Word, and Excel documents to create the test load.
PowerVault 715N test configuration
Dell tested a PowerVault 715N server with the following configuration:
- Processor: Intel® Pentium® III processor at 1 GHz
- Memory: 512 MB SDRAM
- Storage: Four internal Western Digital® 7200 rpm 120 GB hard disk drives (480 GB total)
- Network connectivity: Two on-board Intel 82559 10/100 Mbps Ethernet network interface cards (NICs); the NICs were load balanced during testing
- Operating system: Microsoft Windows Powered operating system based on Windows 2000 Advanced Server
- On-board RAID level: RAID-5
- PCI (Peripheral Component Interconnect) slots: One
Dell created a virtual client environment with four Dell PowerEdgeTM 2550 servers as LoadRunner client drivers. The client drivers ran LoadRunner scripts that loaded virtual clients on the PowerVault 715N to simulate user activity.
The load-testing workload
The test scenario started with approximately 10 GB of data stored on the PowerVault 715N system, which increased to more than 12 GB during a test run. Each run consisted of FTP and Common Internet File System (CIFS) data transactions that emulated a typical user environment for a small- to medium-size company. The data-intensive transactions performed random file modification, deletion, and creation, with a maximum of two-second think times, which were used mainly in the FTP transactions.
Dell used four LoadRunner scripts to test the PowerVault 715N server. The file_copy_modify script copied files from a client to the NAS server and modified several files. The Ftp_copy script executed batch transfers of 64 KB to 1 MB files. The File_copy_delete_modify script executed several copy-delete-modify actions within a 4 MB file folder that held files ranging from 28 KB to 2 MB. The last script executed a combination of the three previously mentioned scripts; each of the script transactions was divided equally in the test run.
Results for PowerVault 715N sizing
The scalability test measured the number of Vusers (simulated user clients) and amount of Windows resources used for each ramp-up of user participation. The transaction response time was set to give an error if it peaked above a 10-second average for CIFS transactions and 50 seconds for FTP transactions. These settings resulted in a more "real-world" type of test environment, simulating peak use time of a medium-size company. Figure 1 , which summarizes the test results, contains the following details:
Figure 1. Scaling for simultaneous users on the PowerVault 715N
- Maximum Resources/Users: Marks the threshold for an acceptable number of simultaneous users and level of performance in an environment similar to the one used in this testing scenario
- Vuser Logins: Indicates the number of clients logged in at a given point in testing
- Win-Resources: Indicates the percentage of Windows resources used during a test run
The Win-Resources line represents data from an actual test run of the PowerVault 715N configured for a small-size company processing higher levels of data with each user ramp-up. The Maximum Resources/Users line indicates the point where the NAS server uses approximately 70 percent of its Windows resources; using this measure as a threshold allows for process spikes and room to scale as resources are depleted during a company's growth.
Figure 1 can help to determine the expected performance of the PowerVault 715N in a small- to medium-size company that creates and modifies data types similar to those used in the test. Companies can use these results to determine or size workload environments requiring entry-level NAS devices, such as the PowerVault 715N server, to store important data from a client group.
However, factors such as drive configuration and data types will affect performance. For example, heavy access of data types such as image files could dramatically affect overall performance and the maximum acceptable number of simultaneous users. The testing discussed in the next section examines the effects of system and network configuration on performance.
Performance of the PowerVault 715N
Dell conducted the performance tests using the Ziff DavisTM NetBench® benchmark tool. The software produces performance data that can be easily plotted to represent overall throughput and average response times. It attempts to re-create the workload of users running file-server applications. The user's workload is simulated by running a sequence of operations, such as "read a file, pause while editing, write the file" followed by a random "make a directory, rename a file, and delete a file." A built-in delay of 5 seconds and a simulated user think time of 2 seconds occur between each operation. These delays let the test closely simulate an actual user's pattern of writing and reading data to and from the server.
Dell ran the NetBench Enterprise Mixed Disk Test to compute the total data throughput achieved given an ever-varying client load. Throughput is the rate at which data is sent back and forth between the PowerVault 715N and its network clients; greater throughput means faster performance. The results are reported in megabits per second (Mbps). If a megabytes per second (MB/sec) value is needed, divide the megabits per second value by 8. The results are plotted in a scalability curve to demonstrate how throughput for the PowerVault 715N scales to incremental client loads.
Dell also used the Enterprise Mixed Disk Test to produce data points that represent average response times-the average time taken by the PowerVault 715N to respond to and complete network client requests at a given client load. Average response times, which are reported in milliseconds, have an inverse relationship to throughput: the higher the number, the slower the performance.
All tests used 16 PowerEdge 1400 servers as network clients to generate network throughput and create the necessary client load. Each client ran the Microsoft Windows 2000 with Service Pack 2 operating system; therefore, only CIFS testing was performed. Hardware configuration for each PowerEdge 1400 server was identical: one 1 GHz processor, 1 GB of memory, and an Intel 8255x-based PCI Ethernet adapter (10/100 Mbps).
Maximum throughput and best average response times
Dell compared the performance between the three currently available models of the PowerVault 715N server:
- 160GB model: Intel Celeron® processor at 850 MHz; 384 MB SDRAM; four 40 GB Integrated Drive Electronics (IDE) hard drives
- 320GB model: Intel Pentium III processor at 1 GHz; 384 MB SDRAM; four 80 GB IDE hard drives
- 480GB model: Intel Pentium III processor at 1 GHz; 512 MB SDRAM; four 120 GB IDE hard drives
These configurations were tested on a segmented 100BaseT network using the on-board NICs (see Figure 2 ).
Figure 2. Network configuration of test environment
As expected, the system with the most memory and the fastest processor, the 480GB model, outperformed the other platforms (see Figures 3 and 4 ). In a RAID-0 configuration, the maximum throughput and average response time for the next-best-performing system, the 360GB model, were almost as good as those for the 480GB model. However, once throughput of the 360GB model began to slow, it did so at a faster rate for both RAID levels.
Figure 3. Throughput and average response time for RAID-0 configurations
Figure 4. Throughput and average response time for RAID-5 configurations
Optimal network configuration
Two potential bottlenecks exist when using the on-board 100BaseT network interface. The first can occur when both on-board NICs are configured to be on the same physical and logical subnet. In this case, the Windows operating system will allow only one NIC to communicate on the network at a time. The operating system has no built-in load-balancing feature that spreads packet transmits and receives across both on-board NICs. When only one NIC is transporting data to the processor, data packets will begin to queue. In turn, the processor will remain idle while waiting for the queue to be processed and will not reach full utilization. To avoid this situation, administrators should separate both on-board NICs onto different subnets.
The other bottleneck can be introduced when the teaming function of the Intel NIC drivers is configured. Using the Fault Tolerant or Load Balancing feature will not double the bandwidth of the on-board 100BaseT network. These features activate only one NIC; the other remains in a wait state for a failure or busy condition. The Load Balancing mode actually uses 5 percent of the processor's time without doubling network bandwidth and can therefore negatively affect performance. Performance degradation can be avoided by configuring the NICs in Link Aggregate mode, if the network switch supports this function. The Link Aggregate mode uses both on-board NICs, creating a bigger pipe through which data packets can be transferred for processing.
Administrators can achieve optimal network performance with two on-board 100BaseT NICs by eliminating one of the bottlenecks. Performance can also be improved by installing a faster NIC; the PowerVault 715N supports an optional Broadcom® NetXtremeTM 10/100/1000 copper Gigabit (CuGb) Ethernet NIC, which can be installed in the 32-bit/3 MHz PCI slot of the storage server.
To determine the optimal network configuration, Dell compared three separate network configurations that included the PowerVault 715N server (480GB model) and the 16 PowerEdge clients. The on-board NICs of the PowerVault 715N were separated onto two different subnets (see Figure 2 ). In the first configuration, only the storage server had a Gigabit Ethernet* card; each network client used a 10/100 Mbps NIC. In the second configuration, all systems were on a Gigabit Ethernet network, and each system (server and clients) had a Gigabit Ethernet network card installed. And in the third configuration, all systems were on a segmented 100BaseT network and used 10/100 Mbps NICs. Figure 5 shows the performance of the three different network configurations.
Figure 5. Throughput for different network and RAID configurations
An indicator of real-world performance
Performance lab testing showed that the PowerVault 715N can achieve almost 240 Mbps under the NetBench stress test. To achieve even better NetBench throughput, administrators should add more processing capacity to the system. Performance monitoring showed that, on every test run, throughput slowed when the processor was at full utilization. When the slowdown occurred, neither the disk subsystem, memory, nor the network were bottlenecks.
A highly network-bound stress test, NetBench exercises a NAS system well beyond real-world conditions. Therefore, the number of clients used in the performance test, along with their associated I/O activity, will not directly correlate to the same number of network clients and associated I/O activity in real-world network environments. The performance results are useful when comparing the PowerVault 715N to other NAS systems in the industry or comparing the performance of different hardware configurations.
The results of the scalability and performance tests can be used to extrapolate real-world performance figures. The scalability test demonstrated that the PowerVault 715N NAS server can provide ample throughput for 130 concurrent users at peak work hours and for workloads similar to the ones tested. In terms of features, performance, and price, the Dell PowerVault 715N is a storage server well suited for a small company or workgroup.
Julia Courtney (firstname.lastname@example.org) is a performance engineer in the Systems Performance and Analysis Lab at Dell. Julia has a B.B.A. in Management Information Systems from the University of Texas at Austin and an M.B.A. in Telecommunications from St. Edward's University. She is a Microsoft Certified Systems Engineer.
Aaron Burns (email@example.com) is a senior solution engineer/ consultant on the Solution Enablement Lab and Technology Showcase team in the Enterprise Systems Group at Dell. He received a B.S. in Computer Science from the University of Texas at San Antonio.