|
Disclaimer: The performance data in the figures on this page were obtained in a controlled laboratory environment using specific performance benchmarks and tools. This information is presented to assist the reader in gaining a better understanding of IBM products. Results obtained in other environments may vary significantly and hence these performance numbers do not predict a specific customer environment.
Objective: This page attempts to show the performance you might expect from the BRMS Application Client when used on various iSeries Models with differing CPW ratings, workloads, and network configurations.
Terms: The following defines some of the acronyms and terms used on this page.
- Gb refers to gigabit or 1,073,741,824 bits
- GB refers to gigabytes or 1,073,741,824 bytes
- Gbps refers to gigabits per second
- IOA is an Input/Output adapter
- IOP is an Input/Output processor
- Mb refers to megabit or 1,048,576 bits
- MB refers to megabytes or 1,048,576 bytes
- Mbps refers to megabits per second
- MHz refers to megahertz or 1,048,576 cycles
- Average network transfer rate refers to the rate TSM received
each block of saved data from BRMS across the network.
- Effective transfer rate refers to the calculated rate for the complete
save of the library and includes the time for BRMS and the operating
system to set up the save and to prepared each block for transfer to TSM.
iSeries Clients: The following describes the iSeries systems used to measure the BRMS Application Client performance.
- System_A - Model 170 - OS/400 V5R2
- Processor feature 2386, 512 MB memory
- #2809 PCI WAN/LAN/Workstaion IOP (Slot E10)(Shared)
- #4838 PCI 100/10 Mbps Ethernet IOA (Slot E08)
- System_B - Model 170 - OS/400 V5R2
- Processor feature 2388, 448 MB memory
- #2809 PCI WAN/LAN/Workstaion IOP (Slot E10)(Shared)
- #4838 PCI 100/10 Mbps Ethernet IOA (Slot E08)
- System_C - Model 720 - OS/400 V5R1
- Processor feature 2392, 4096 MB memory
- #4838 PCI 100/10 Mbps Ethernet IOA (Slot C01)
- System_D - Model 820 - OS/400 V5R2 (PTF MF28304)
- Processor feature 2398, 4096 MB memory,
- #2843 PCI IOP (Slot C03)(Dedicated)
- #2743 PCI 1Gbps Ethernet PCI IOA (Slot C02)
- #4838 PCI 100/10 Mbps Ethernet IOA (Slot C01)
Note: Refer to Chapter 5, Communications in the iSeries Performance Capabilites Reference book for tips on configuring TCP/IP for optimum performance. This book can be found on the Internet at the following URL:
http://publib.boulder.ibm.com/pubs/pdfs/as400/V4R5PDF/as4ppcp5.pdf (PDF, 1.1MB)
Network: The following describes the network used between the iSeries systems and the TSM server.
- The 100 Mb Ethernet was a private network consisting of an IBM 8223 Fast Ethernet Network Hub, the iSeries clients and the TSM/AIX server.
- The 1 Gb Ethernet was a direct connection between the iSeries and the TSM server using a pair of Gigabit Ethernet-SX PCI Adapters.
- All 100 Mb Ethernet testing was performed using half-duplex mode.
- All testing was performed using TCP/IP.
TSM server: The following describes the TSM server used.
- IBM RS/6000 43P Model 150, 375 MHz processor, 1024 MB memory
- IBM 10/100Mb Ethernet PCI Adapter
- Gigabit Ethernet-SX PCI Adapter (using jumbo frames)
- AIX Version 4.3.3
- Tivoli Storage Manager Version 5.1.0
The server storage was set up as follows:
- a 3 GB Database using 3 volumes of 1 GB each
- a 2 GB Recovery Log using 2 volumes of 1 GB each
- a 14 GB Storage Pool using 7 volumes of 2 GB each
Method: The following describes the method used to generate the the performance data.
- A series of 25 libraries of increasing size ranging from 1 MB to 1 GB were created on the client.
- These libraries were saved to the TSM server sequentially from smallest to largest.
- Each series was saved four times and an average of the effective transfer rate for these four saves is presented in the results below.
- The effective transfer rates were calculated for each library by dividing the number of bytes transfered during the save by the aggregate save time. The aggregate save time is the time it took BRMS to setup for each library save, plus the time it took the operating system to save the library, plus the time it took BRMS to perform post processing after the save.
- Measurements were taken and results tabulated when restoring these libraries from the TSM server.
- The save and restore measurements were performed on all clients using a 100 Mb Ethernet connection to the TSM server.
- These save and restore measurements were repeated for the Model 820 and Model 52M clients using a 1 Gb Ethernet connection to the TSM server.
- The save scenario was repeated simulating a network/TSM server of infinite capacity. This is intended to show how fast the iSeries client and BRMS can provide save data to the Tivoli Storage Manager Application Programming Interface (TSM APIs) and TSM server. This was accomplished by modifying the BRMS programs such that instead of calling the TSM APIs to transfer the save data, the save data was written to memory.
- A similar save scenario using a series of 25 directories of increasing size ranging from 1 MB to 1 GB was also performed using the infinite capacity scenario to identify any performance differences which might exist between library saves and directory saves.
Results of the performance benchmarking follows.
Results: The following figures document the results of the save and restore performance benchmarking of the BRMS client on several iSeries models and networks.
Each figure represents an iSeries model and network type. The figures are presented in order of increasing CPW performance capability of the iSeries model.
Each figure shows the effective transfer rates of a save and restore across the network, a save of the libraries assuming an infinite network, and a save of directories assuming an infinite network.
Below each figure is a summary of the typical system activity at the time of the save, some observation about the data, and a link to a table containing the data used in the figure.
Model 170 (CPW 460) - 100Mb Ethernet
Figure 1: System_A
View data
System activity
- The system was running slightly over 165 active jobs at the time of the saves.
- CPU utilization was slightly over 25 percent for all active jobs for the duration of the saves.
- The average network transfer rate for all saves was 15.4 GB/hour on the 100Mb Ethernet.
- The maximum effective transfer rate was approximately 8 GB/hour for 1GB libraries which was limited by the low CPW and the relatively low average network transfer rate.
Model 170 (CPW 1090) - 100Mb Ethernet
Figure 2: System_B
View data
System activity
- The system was running slightly over 180 active jobs at the time of the saves.
- CPU utilization was slightly over 20 percent for all active jobs for the duration of the saves.
- The average network transfer rate for all saves was 10.3 GB/hour on the 100Mb Ethernet.
- The maximum effective transfer rate was approximately 13 GB/hour for 1GB libraries which was limited by the low average network transfer rate.
Model 720 (CPW 1600) - 100Mb Ethernet
Figure 3: System_C
View data
System activity
- The system was running slightly over 230 active jobs at the time of the saves.
- Utilization of all jobs was under 10% for all active jobs for the duration of the saves.
- The average network transfer rate for all saves was 17.2 GB/hour on the 100Mb Ethernet.
- The maximum effective transfer rate was approximately 14 GB/hour for 1GB libraries which was limited by the low average network transfer rate.
Model 820 (CPW 3200) - 100Mb Ethernet
Figure 4: System_D
View data
System activity
- The system was running slightly over 200 active jobs at the time of the saves.
- CPU utilization was under 10 percent for all active jobs for the duration of the saves.
- The average network transfer rate for all saves was 27 GB/hour on the 100Mb Ethernet.
- The maximum effective transfer rate was approximately 27 GB/hour for 1GB libraries which was limited by the network capability.
Model 820 (CPW 3200) - 1Gb Ethernet
Figure 5: System_D
View data
System activity
- The system was running slightly over 190 active jobs at the time of the save.
- CPU utilization was under 10 percent for all active jobs for the duration of the backup.
- The average network transfer rate for all saves was 59 GB/hour on the 1Gb Ethernet.
- The maximum effective transfer rate was approximately 32 GB/hour for 1GB libraries.
Observations:
Performance appears to be proportional to the CPW available to the backup job.
Performance appears to be proportional to the available network bandwidth.
Performance on older iSeries models may be limited if the communications IOA is attached to a shared multi-function IOP.
CPU utilization typically increases proportionally with the effective network data rate.
Performance for saves of directories and files appears typically to be less than saves of libraries and objects.
Performance of large objects is always better than saves of small objects.
Performance for restores is typically much less than saves.
Rules of Thumb:
Use a 1Gb Ethernet connection between the iSeries system and the TSM whenever possible. This supports a larger maximum frame size of 8996 bytes compared to the 1496 byte maximum on a 100Mb Ethernet connection and provides better performance.
Use a dedicated IOP for the communications IOA if this is supported by the iSeries hardware configuration.
If you are running on OS/400 Version 5 Release 1 Modification 0, install PTF MF28304 to assure the best level of operating system performance.
If larger effective data rates for the backup are required and the limitation appears to be a low CPW rating for the system or IOA performance, consider concurrent backup jobs or multiple networks if CPU utilization is relatively low.
| 
