NERSC logo National Energy Research Scientific Computing Center
  A DOE Office of Science User Facility
  at Lawrence Berkeley National Laboratory
  3PARdata S400 Storage I/O Performance Results

3PARdata S400 Storage I/O Performance Results

1. Overview

This report documents the timing results of the performance tests done on the 3PAR InServ S400 Storage Server. The primary focus of the performance test was to measure the maximal single-port, single-node, and single-system performance (bandwidth in MB/second). The tests included single-stream and multi-stream tests with different file sizes and I/O sizes. The evaluation period was from August 2003 to October 2003.

2. Test Configuration

We used NERSC MPTIO and PIORAW benchmarks to test read and write small (in-cache) and large (out-of-cache) files. Except for the file system tests, all the tests were performed using the raw device interface (e.g., /dev/raw/raw1) to avoid the use of host system buffer cache.

The block sizes used in the tests include: 1 KB, 4 KB, 16 KB, 64 KB, 256 KB, 1 MB, 4 MB, and 16 MB.

For multi-stream parallel I/O tests, the aggregate bandwidth was calculated as the total number of bytes read or written divided by the longest elapsed time.

All tests were run on a quiet system. There were no other activities, neither on the clients nor on the 3PARdata storage when the tests were running.

2.1 Linux Host Configuration

The test host had the following configuration:

·       Dual 2.2 GHz Xeon P4 processors, SuperMicro motherboard

·       2 GB 133 MHz ECC memory

·       RedHat 7.3 with 2.4.18-10smp kernel

·       Qlogic QLA2340 2 Gb HBA with qlogic driver v6.1b2 (with a patch that fixed the bounce buffer problem)

2.2 3PAR Storage Configuration

The 3PARdata InServ S400 Storage Server consisted of:

·       Four controller nodes, each with 4 GB memory

·       Each controller node had six PCI slots for host connectivity and back drive chassis

·       Only four PCI slots were populated, two for front-end host connections and two for backend drive chassis connections

·       Each slot sits dual 2 Gb FC ports


3. Performance Results

3.1 Single FC Port Performance

Objective: To measure S400 single FC port performance.

Test Setup:

·       Client: Linux 2.4.18-10smp, Dual 2.2 GHz Xeon P4 with QLA2340 HBA

·       Storage: 3PARdata

·       # of FC Ports and Speed: one 2 Gb FC port

·       LUN Configuration: One RAID-5 or RAID-1 LUN

·       I/O Size: 1 KB to 16 MB

·       # of Client: 1

·       # of I/O Processes per Client: 4

·       Benchmark: PIORAW

 

RAID-5 Test Result:

·       LUN Configuration: RAID-5 LUN

·       File Size: 128 MB

 

 

 

 

 

 

 

 

 

 

 


 


Detail Results: 1-port performance (R5)

I/O Size

Write

Read

1 KB

3.28 MB/s

6.65 MB/s

4 KB

13.82 MB/s

24.49 MB/s

16 KB

52.86 MB/s

84.94 MB/s

64 KB

147.47 MB/s

189.14 MB/s

256 KB

191.00 MB/s

195.88 MB/s

1 MB

194.87 MB/s

197.67 MB/s

4 MB

194.66 MB/s

197.61 MB/s

16 MB

195.01 MB/s

197.61 MB/s

 

RAID-1 Test Result:

·       LUN Configuration: RAID-1 LUN

·       File Size: 256 MB

 

 

 

 

 

 

 

 

 

Detail Results: 1-port performance (R1)

I/O Size

Write

Read

1 KB

3.31 MB/s

6.77 MB/s

4 KB

13.50 MB/s

24.43 MB/s

16 KB

51.73 MB/s

84.44 MB/s

64 KB

145.25 MB/s

188.96 MB/s

256 KB

187.57 MB/s

195.88 MB/s

1 MB

192.18 MB/s

197.63 MB/s

4 MB

193.25 MB/s

197.62 MB/s

16 MB

190.51 MB/s

197.64 MB/s

 

Result Summary:

The above results show 3PARdata's S400 single FC port performance for cached I/O's on both a RAID-1 LUN and a RAID-5 LUN. The results were the sequential I/O performance, for reading and writing of a 128 MB file, using different I/O sizes. Since the memory cache on each of the InServ nodes is 8 GB, the content of the 128 MB region can be entirely cached in the InServ controller node. With the file content cached, the performance would show the best possible transfer rate between a host and a FC port on the storage controller.

The best single-port performance was about 195 MB/s for writes and 197 MB/s for reads. Since the entire file could be cached in the controller, we expected to see similar performance results on RAID-1 and RAID-5.

3.2 Single-Controller Performance

Objective: To measure S400 single-controller performance.

Test Setup:

·       Clients: Linux 2.4.18-10smp, Dual 2.2 GHz Xeon P4 with QLA2340 HBA

·       Storage: 3PARdata

·       # of FC Ports and Speed: 4 2 Gb FC ports

·       LUN Configuration: 1 RAID-5 LUN

·       File Size: 128 MB and 16 GB

·       I/O Size: 1 KB to 16 MB

·       # of Clients: 4

·       # of I/O processes per host: 4

·       Benchmark: PIORAW

 

In-Cache I/O Test Result: