Workload Characterization for VM and process in Non-virtualized Environment

We want to characterize or understand the behavior of a virtual machine (VM). Is it a CPU-intensive or I/O-intensive VM?

We are going to use the following tools such as xentop and pidstat. 

Let's see which metrics would be appropriate to use to identify the characteristics of VM. 

VIRTUALIZED ENVIRONMENT

<<VM Monitor Test Framework>>

Monitoring

1. XenTop

When do we have to monitor dom0 and domU ??

source is available under tools/xenstat/xentop/xentop.c

2. XenTop

># xentop -b -d 1 -i 2

NON-VIRTUALIZED ENVIRONMENT

<<PS Monitor Test Framework>>

Monitoring

1. top/ps/pidstat - CPU, Memory

* fork로 생성된 child processes에 대해서는 어떻게 되는가? 여기에도 포함이 되는지? 

># ps -p <pid> -o pid,pcpu,pmem,comm

 PID %CPU %MEM COMMAND

19328  100  0.0        loop

It seems like pidstat shows memory access as well (?) and since this command shows the stats based on the name of process so I think I should rather use this instead of top or ps command. 

># pidstat -C stress 1

01:56:17 PM       PID   %user %system    %CPU   CPU  Command

01:56:18 PM     20849    5.94   93.07   99.01     3  stress

01:56:18 PM     20850    3.96   97.03  100.99     0  stress

># pidstat -r -C stress 1

01:56:37 PM       PID  minflt/s  majflt/s     VSZ    RSS   %MEM  Command

01:56:38 PM     20849 223570.30      0.00  268388  75420   1.05  stress

01:56:38 PM     20850 223525.74      0.00  268388  75420   1.05  stress

2. pidstat/iotop - Disk

># apt-get install sysstat

># pidstat -d -p 19744

or 

># pidstat -C stress -d 1

01:43:37 PM       PID   kB_rd/s   kB_wr/s kB_ccwr/s  Command

01:43:38 PM     19744      0.00      0.00  43089.11  stress

01:43:39 PM     19744      0.00      0.00  48640.00  stress

01:43:40 PM     19744      0.00      0.00 139405.94  stress

3. nethogs - Network

># nethogs -t | grep 18644

Waiting for first packet to arrive (see sourceforge.net bug 1019381)

sshd: ykoh@pts/0/18644/1001 1928 216

sshd: ykoh@pts/0/18644/1001 3696 324

sshd: ykoh@pts/0/18644/1001 5624 540

sshd: ykoh@pts/0/18644/1001 7392 702

Benchmarking

<<VM Monitor Test Framework>>

Preparation

* do.sh script executes given commands on all VMs

1. SPECcpu

Build the spec benchmarks

># scripts/do.sh runspec --config=vm.cfg --action=build --tune=base all

># scripts/do.sh runspec --config=vm.cfg --size=test --tune=base --noreportable --iterations=1 all

1. Measuring CPU/Memory Performance with SPECcpu

># 

2. Measuring Disk Performance with iozone and bonnie++

3. Measuring Network Performance with netperf and httperf

4. Measuring with Application Benchmark - POV-Ray 

* This is part of SPECcpu so just skip this. 

References

1. Xen Monitoring Tools and Techniques, http://www.virtuatopia.com/index.php/Xen_Monitoring_Tools_and_Techniques

2. XenMon QoS Monitoring and Performance Profiling Tool, HP Development Company, 2005

3. Xen performance Tools, http://xen.1045712.n5.nabble.com/summary-xen-performance-tools-td2544469.html

4. Profiling and Benchmarking Under Xen, http://book.xen.prgmr.com/mediawiki/index.php/Chapter_10:_Profiling_and_Benchmarking_Under_Xen

5.