Veritas NetBackup™ Appliance Capacity Planning and Performance Tuning Guide
- About this Guide
- Section I. Capacity planning
- Section II. Best Practices
- Section III. Performance tuning
- Section IV. Quick reference to Capacity planning and Performance tuning
About network monitoring and tuning
Table: Network Statistics (collected using ifstat 5) shows the network statistics collected with ifstat when the system was running 120 concurrent 98% deduplication backup streams.
Table: Network Statistics (collected using ifstat 5)
eth5 | eth4 | eth6 | eth7 | bond0 | ||||||
---|---|---|---|---|---|---|---|---|---|---|
HH:MM:SS | in | out | in | out | in | out | in | out | in | out |
17:26:21 | 564730.9 | 1302.0 | 1160000.0 | 2140.2 | 559536.6 | 1621.4 | 168878.4 | 1153.4 | 4710000.0 | 12498.8 |
17:26:26 | 460546.3 | 728.0 | 1150000.0 | 698.9 | 459263.6 | 1330.3 | 381489.9 | 1142.0 | 4770000.0 | 11797.7 |
17:26:31 | 227735.0 | 537.6 | 1150000.0 | 724.0 | 445882.1 | 951.5 | 444105.8 | 1069.2 | 4680000.0 | 9971.4 |
17:26:36 | 374109.8 | 962.3 | 889411.5 | 863.5 | 421187.4 | 450.1 | 423359.6 | 1091.7 | 4740000.0 | 8593.0 |
17:26:41 | 413063.0 | 826.2 | 827849.0 | 765.3 | 415254.8 | 454.0 | 414780.8 | 758.3 | 4710000.0 | 6980.5 |
17:26:46 | 416389.0 | 802.9 | 822851.8 | 309.1 | 413860.3 | 425.2 | 409673.9 | 393.7 | 4690000.0 | 5694.1 |
17:26:51 | 413397.2 | 644.5 | 866869.2 | 587.3 | 343333.5 | 439.7 | 407649.9 | 343.9 | 4650000.0 | 5744.9 |
17:26:56 | 428205.4 | 510.6 | 1150000.0 | 2804.2 | 224508.3 | 952.0 | 427518.5 | 592.0 | 4550000.0 | 9509.6 |
17:27:01 | 474236.7 | 1040.3 | 1090000.0 | 2321.1 | 473665.8 | 1309.5 | 472919.9 | 832.9 | 4680000.0 | 12125.6 |
Note:
Some of the columns from the output have been removed to simplify the display.
The above table showed the network throughput on the appliance of selected NIC ports. The appliance has 10 x 10 Gbps NICs bounded with the default balance-alb (adaptive load balance) mode. Under each NIC interface, the in column displays the kilobyte received per second from the network, while out column displays the kilobytes transmitted out per second. Eth4 is the primary slave which, in a couple 48 port switch models, tends to have a higher workload than the other slave NIC port. The data points highlighted in bold indicate that the NIC port is near capacity. With this condition, there is a possibility for packet loss. You can run the command "netstat - s |grep retransmit", periodically to check for the number of retransmits. Nonzero value initially is fine, but if you see that this value increases, then there may be a network congestion issue.
The standard tuning recommendations are:
Add or enable additional NIC ports
Change the bonding mode, especially when all other ports, except the primary slave, still have plenty of extra bandwidth. The default balance-alb mode can sometimes cause this problem on some switch models. Our internal test showed 802.3ad (LACP) bond mode can solve this problem.