• No results found

Try to ping the default gateway to check the HEA configuration

In document AP254XINST (Page 144-150)

EXempty Exercise 6. I/O device virtualization performance and tuning

__ 59. Try to ping the default gateway to check the HEA configuration

__ 60. In the following steps, you will start a network load and monitor the statistics of the VIOS partition’s shared Ethernet adapter.

__ 61. Log in to the partition with the HEA port configured. As the root user, check the status of the I/O Completion Ports pseudo device with the following command:

# lsdev -l iocp0

If the device is marked as Defined, then activate the device permanently with the following command sequence:

# chdev -l iocp0 -a autoconfig=available

# mkdev -l iocp0

__ 62. As the root user, verify the netperf server is running. To determine if the netperf server is running, use the netstat command to verify that the default port 12865 has LISTEN for its status as shown below.

# netstat -an | grep 12865

tcp4 0 0 *.12865 *.* LISTEN If netperf is not running, start it by issuing the following command:

# /home/an31/ex6/start_netserver

__ 63. Repeat the previous two steps on your first LPAR with the virtual Ethernet adapter configured.

__ 64. As the root user on your second partition (the one with the HEA port configured), use the tcp_rr.sh script located in the directory /home/an31/ex6 to generate a network load to the other logical partition.

Syntax of the tcp_rr.sh script:

# ./tcp_rr.sh <remote host IP> <sessions> <duration>

The tcp_rr.sh script requires three arguments: the IP address of the remote partition, the number of sessions, and the duration of the test. Start the command with 10 sessions and a duration of 300.

Here is an example command. Substitute 9.47.88.153 with the actual IP address of your remote logical partition:

# ./tcp_rr.sh 9.47.88.153 10 300

Course materials may not be reproduced in whole or in part

© Copyright IBM Corp. 2010, 2013 Exercise 6. I/O device virtualization performance and tuning 6-27

EXempty » There are a few ways of using the netstat command to see packet statistics. Here is one example, followed by a sample output. Substitute en0 with the device name for the virtual Ethernet adapter on your logical partition. __ 66. From the padmin CLI on the VIOS partition, use the netstat command to list the

packet flow going through the shared Ethernet adapter device. Do you see any IP packets? Why or why not?

$ netstat -stats 2

» Here is the output of the command netstat -stats 2:

$ netstat -stats 2 » Notice the statistics on the left side are related to the interface en3. The statistics on the

right side of the output are the total statistics on the partition. There might be a small number of packets received and transmitted, however they are likely the packets used to transmit output to the terminal you have used to log in (assuming you have not used the console window from the HMC GUI application).

» You might have expected to see the total number of packets received and transmitted (displayed on the right side of the output) be at least the same as the statistics being shown on your logical partition (around 52000 packets). However, remember that the shared Ethernet adapter is a layer 2 bridge device. It is functioning at the Ethernet frame level, not at the IP packet level, which is where netstat reports statistics. The way to see the amount of packets going through the shared Ethernet adapter is to use the enstat entX command on the shared Ethernet adapter logical device name.

__ 67. On the VIOS partition, try to list the packet flow going through the physical and the virtual adapters associated with the shared Ethernet adapter, using the netstat or enstat command. What are the results?

$ entstat entX

$ netstat -stats 2

» Use lsdev to list all of the Ethernet adapters on the VIOS partition.

$ lsdev | grep ent ent0 Available 10/100 Mbps Ethernet PCI Adapter II (1410ff01) ent1 Available Virtual I/O Ethernet Adapter (l-lan) ent2 Available Virtual I/O Ethernet Adapter (l-lan) ent3 Available Shared Ethernet Adapter

» If you are following along with the examples in this exercise, the real physical adapter is ent0 and the associated virtual adapter is ent2. You can see this on your system from the output of the lsdev -dev ent3 -attr command. The ent2 adapter is virtual, and it is configured for access to the VIOS partition.

» It is not possible using enstat or netstat to directly display the statistics of the virtual and physical devices associated with the shared Ethernet adapter. Here is an example of error messages when trying enstat:

$ entstat -all ent2

entstat: 0909-003 Unable to connect to device ent1, errno = 19

$ entstat -all ent0

entstat: 0909-003 Unable to connect to device ent0, errno = 19 __ 68. A way to see the shared Ethernet adapter statistics is to list the Ethernet device

driver and devices statistics of the shared Ethernet adapter itself. Examine the statistics for the adapters in the Shared Ethernet Adapter section of the output.

Execute this command multiple times to see the statistics updating.

From the padmin CLI on the VIOS partition:

$ entstat -all entX | more

Replace entX with the name of the shared Ethernet adapter device. The output of the command can be quite large. If you only want to view the real and virtual side statistics for packet counts, try passing the output to grep for the paragraph titled Statistics for adapters in the shared Ethernet adapter entX.

For example:

Course materials may not be reproduced in whole or in part

© Copyright IBM Corp. 2010, 2013 Exercise 6. I/O device virtualization performance and tuning 6-29

EXempty » Here is an output example of the command:

$ entstat -all ent3 | grep -p SEA

---Statistics for adapters in the Shared Ethernet Adapter ent3 Memory allocation failures: 0 ICMP error packets sent: 0

Non IP packets larger than MTU: 0 Thread queue overflow packets: 0

---Real Adapter: ent0

__ 69. Another way to check the statistics of the shared Ethernet adapter is using the seastat command. The seastat command generates a per client view of the shared Ethernet adapter statistics. To gather network statistics at this level of detail, advanced accounting should be enabled on the shared Ethernet adapter to provide additional information about its network traffic.

__ a. Use the chdev command to enable the advanced accounting on the SEA.

$ chdev -dev entX -attr accounting=enabled

Replace entX with your shared Ethernet adapter device name.

__ b. Invoke the seastat command to display the shared Ethernet adapter statistics per client logical partition. Check for the transmit and receive packets number increasing.

$ seastat -d entX

» Here is an output example of the command:

$ seastat -d ent3

===============================================================

Advanced Statistics for SEA Device Name: ent3

================================================================

MAC: 56:ED:E7:C9:03:0B ---VLAN: None

VLAN Priority: None IP: 10.6.112.42

Transmit Statistics: Receive Statistics:

- ---Packets: 8199931 ---Packets: 8199380 Bytes: 483824638 Bytes: 491974792

__ 70. Stop any monitoring commands that are running in any of your partitions. Stop the tcp_rr.sh script with Ctrl C.

Tuning the shared Ethernet adapter

There are two things to consider when tuning a configuration that uses a shared Ethernet adapter.

The first relates to the configuration of the TCP/IP attributes on the partitions that are using the shared Ethernet adapter. These attributes might be system wide on the partition, or they might be ISNO settings for the particular virtual Ethernet interface. The normal TCP/IP tuning must be performed to maximize the throughput available in a virtual Ethernet or shared Ethernet adapter configuration, and this is beyond the scope of this course.

The second thing to consider is the shared Ethernet adapter device itself. For the shared Ethernet adapter, the only thing that can really be tuned is the CPU capacity entitlement of the virtual I/O server partition. If the virtual I/O server partition is CPU constrained, then the throughput of the shared Ethernet adapter might be reduced. This is what you will explore

Course materials may not be reproduced in whole or in part

© Copyright IBM Corp. 2010, 2013 Exercise 6. I/O device virtualization performance and tuning 6-31

EXempty __ 71. Use the HMC GUI to dynamically configure both of your AIX logical partitions as uncapped with an uncapped weight value to 128. You can leave the assigned processing units at the current setting.

» Select a partition and run the Dynamic Logical Partitioning > Processor > Add or Remove task. Make sure the uncapped checkbox is checked and the weight value is 128. Repeat for the other AIX partition.

__ 72. Use the HMC GUI to check that your VIOS partition has the following CPU configuration:

• Assigned Processing units: 0.10 • Assigned Virtual Processors: 1 • Partition is Capped

If needed, dynamically change your VIOS partition configuration using the HMC GUI.

» Using the HMC GUI, from the Navigation area on the left, select your server, then your assigned VIOS partition. From the Tasks menu, select Dynamic Logical Partitioning

> Processor > Add or Remove.

» Reconfigure the partition, and then click OK to implement the change.

__ 73. On the VIOS partition that you documented previously with the primary SEA configuration, monitor the CPU consumed (physc value) using the viostat command.

# viostat -tty 2

__ 74. As the root user on your second AIX partition (the partition with the HEA port configured), use the ftpload.sh script located in the /home/an31/ex6 directory to generate a network load between your logical partitions.

Syntax of the ftpload.sh script:

# ./ftpload.sh <remote host IP> <user> <password>

Here is an example command. Substitute abcxyz with the actual password for the root account, and substitute 9.47.88.153 with the actual IP address of your second logical partition:

# ./ftpload.sh 9.47.88.153 root abcxyz

» Here is an example of the output from the script when the CE of the VIOS partition is

314572800 bytes sent in 47.82 seconds (6424 Kbytes/s) local: | dd if=/dev/zero bs=1m count=300 remote: /dev/null 221 Goodbye.

__ 75. Convert the throughput value reported by the ftpload.sh script from Kbytes/s into

In document AP254XINST (Page 144-150)