언제나 영원히

정말 오랜만에,, HPC Server 2008 자료를 검색하다가,,
HPC Cluster 성능 측정 툴을 발견했습니다.
Lizard = Linpack Wizard 를 합친 말로 보입니다.

그동안, www.netlib.org 에서 source를 받아다가,,
컴파일해서 실행했었는데,,, 패키지로 제공하는 군요..
예전 홈페이지에 관련 자료를 올렸었는데요,, 아래 링크에서 확인하세요..
http://www.wccs.co.kr/bbs/view.php?id=QNA&no=19
http://www.wccs.co.kr/bbs/view.php?id=PDS&no=13

다시 Lizard로 돌아와서...
Microsoft HPC Pack 2008 Tool Pack 에 가면,,
lizard.msi 파일을 다운로드 받을 수 있습니다.

Install Process

Overview

This release of the HPC Pack 2008 Tool Pack contains 1 tool.

Lizard - Lizard helps you to determine the computational performance and efficiency that can be achieved by your HPC Pack 2008-based / Windows HPC Server 2008 cluster. It calculates and reports a peak performance value for your HPC cluster in billions of floating-point operations per second (GFLOPS), and a percentage value for the efficiency that was achieved at peak performance. After running Lizard, you can review the performance and efficiency results that were obtained, and optionally save these results and the parameters that were used to achieve them to a file on your computer.

Note: This utility includes parts of the Intel Math Kernel Library. For more information about the Intel Math Kernel Library, please visit http://www.intel.com/cd/software/products/asmo-na/eng/307757.htm

1. 설치를 마친 후 Lizard 를 시작한 화면입니다.

2. Tuning Mode 선택 화면입니다. Express Tuning 과 Advanced Tuning 이 있군요..

3. 가상머신에서 실행해서인지 CPU 특성을 자동으로 찾지 못하는 군요..
Dual Core 이상에서는 1 clock 에 4번, 그 이전 CPU 에서는 1 clock 에 2번 계산합니다.
아래 그림의 선택은 잘못 했네요.. 4로 입력해야 하는데,, 2로 입력했습니다.

4. Integrity Test를 위해서는 4대가 필요하다고 하는데,, 테스트 환경이라 1대로 진행합니다.

5. 뭔가 진행하면서,, Consistency 를 확인하는군요..

6. 실제.. xhpl 프로그램을 실행하면서,, tuning을 진행하는 단계입니다.
 프로그램 크기(N), Block Size (NB), 프로세스 수 (PxQ), 기타 파라미터를 변경하면서,,
 성능을 측정하고 있습니다. 매번 숫자 변경하고,, job submit 하던 과정이 바로 진행되고,,
 그래프로 보여지니까 확실히 편하네요..
 아까, 3번 설정에서 4로 입력해야 하는데,, 2로 입력하니까,,
 효율이 100%를 넘어섰네요..

·         Advanced Deployment Operations in Windows HPC Server 2008
(http://technet.microsoft.com/en-us/library/dd772582.aspx)

·         Understanding Job Validation in Windows HPC Server 2008
(http://technet.microsoft.com/en-us/library/dd772585.aspx)

Service Console에서도 Virtual Machine Online Backup이 가능할까?
앞서 vcb관련 내용을 포스트 했는데,,
여기서 설명하는 내용은 vmdk 파일 자체를 백업 받는 방법입니다.

ESX 2.5.x 인 경우

1. vmAddRedo.pl     # Snapshot 생성
2. vmkfstools -i source.vmdk /vmfs/backup/target.vmdk
3. vmCommit.pl       # Snapshot 제거

# vmware-cmd -l
/root/vmware/win2k3/win2k3.vmx
/root/vmware/win2k/win2k.vmx

# vmAddRedo.pl -c /root/vmware/win2k3/win2k3.vmx
vmAddRedo: VM config file is '/root/vmware/win2k3/win2k3.vmx'

vmAddRedo: Disks found that are in use:
vmAddRedo: Disk found: scsi0:0 (localvmfs:win2k3.vmdk)
vmAddRedo:    1: scsi0:0
vmAddRedo: Creating the RedoLogs for cfg : /root/vmware/win2k3/win2k3.vmx
vmAddRedo:    scsi0:0:Adding REDO
vmAddRedo: diskname :/vmfs/localvmfs/win2k3.vmdk

# grep vmdk /root/vmware/win2k3/win2k3.vmx
scsi0:0.name = "localvmfs:win2k3.vmdk"

# vmkfstools -i /vmfs/localvmfs/win2k3.vmdk /vmfs/backup/win2k3.vmdk

# vmCommit.pl -c /root/vmware/win2k3/win2k3.vmx

vmCommit: VM config file is '/root/vmware/win2k3/win2k3.vmx'

vmCommit: Disks found that are in use:
vmCommit: Disk found: scsi0:0 (localvmfs:win2k3.vmdk)
vmCommit:    1: scsi0:0
vmCommit: Backing up the cfg : /root/vmware/win2k3/win2k3.vmx

vmCommit: scsi0:0: Adding stacked REDO.REDO
vmCommitRedo: diskname :/vmfs/localvmfs/win2k3.vmdk
vmCommit: scsi0:0: Committing REDO
Please wait, committing disks...
vmCommit: scsi0:0: Committing REDO.REDO


VI3 인 경우

1. vmware-cmd ${vm} createsnapshot ${vmname}-snapshot
2. vmkfstools -i source.vmdk /vmfs/volumes/backup/target.vmdk
3. vmware-cmd ${vm} removesnapshots

아래 스크립트는 참고만 하시고,, 그대로 사용하시면 안됩니다.

osbackup.sh

#!/bin/sh
logfile="/tmp/backuplog"

echolog() {
    vmname=$1
    msg=$2
    now=$(date "+%b %d %T")
    echo "${now} ${vmname} # ${msg}"
    echo "${now} ${vmname} # ${msg}" >> ${logfile}
}

/usr/bin/vmware-cmd -l | while read vm
## cat /tmp/vms | while read vm
do
    vmname=$(/usr/bin/vmware-cmd ${vm} getconfig displayname | awk '{print $3}')
    echolog ${vmname} "vmname = ${vmname}"

    echolog ${vmname} "create snapshot"
    if [ $(/usr/bin/vmware-cmd ${vm} createsnapshot ${vmname}-snapshot | awk '{print $3}') == 1 ]
    then
        echolog ${vmname} "snapshot is successfully created"
    else
        echolog ${vmname} "snapshot creation has an error !!!!"
    fi

    echolog ${vmname} "verify snapshot creation"
    if [ $(/usr/bin/vmware-cmd ${vm} hassnapshot | awk '{print $3}') == 1 ]
    then
        echolog ${vmname} "snapshot is successfully detected"
   
        case ${vmname} in
          test1 )
          files="/vmfs/volumes/lovmfs/test1/test1.vmdk"
          ;;
          test2 )
          files="/vmfs/volumes/lovmfs/test2/test2.vmdk"
         ;;
        esac
  
        echolog ${vmname} "backup image"
        for eachfile in ${files}
        do
            fromfile=${eachfile}
            tofile=$(dirname ${eachfile})/BK/$(basename ${eachfile})
            if [ -d $(dirname ${tofile}) ]
            then
                echolog ${vmname} "backup directory already exist"
            else
                echolog ${vmname} "backup directory does not exist, make it"
                mkdir -p $(dirname ${tofile})
            fi
            vmkfstools -i ${fromfile} ${tofile}
            diff ${fromfile} ${tofile}
            if [ $? == 0 ]
            then
                echolog ${vmname} "${fromfile} is successfully copied to ${tofile}"
            else
                echolog ${vmname} "${fromfile} copy has an error !!!!"
            fi
        done
   
        echolog ${vmname} "delete snapshot"
        if [ $(/usr/bin/vmware-cmd ${vm} removesnapshots | awk '{print $3}') == 1 ]
        then
            echolog ${vmname} "snapshot is successfully removed"
        else
            echolog ${vmname} "snapshot removing has an error !!!!"
        fi
    else
        echolog ${vmname} "snapshot does not exist !!!!"
    fi
    echo -n $(date "+%b %d %T") ${vmname} "# "
    /usr/bin/vmware-cmd ${vm} hassnapshot
    ;;
done