Xen GPU Cluster Practice

實作一: 如何將Dom0上的顯示卡資源分配給DomU

Hardware

Machine	Dell OptiPlex 755
Node	1 node
CPU	Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz
Memory	6GB
Storage	160GB
Video Card	NVIDIA GeForce 9800GT 1GB

Software

OS#1	Ubuntu 8.10 with Kernel: 2.6.28 x86_64 (non-xen-patched kernel)
OS#2	Ubuntu 8.10 with Kernel: 2.6.22-9 x86_64 (Xen-3.3.1+Lustre patched kernel)

步驟一: 連線到遠端主機

# 以下兩種連線方式擇一使用.
rider@cloud:~$ ssh 140.xxx.xxx.xxx
rider@cloud:~$ vncviewer 140.xxx.xxx.xxx

步驟二: 產生一台虛擬機器來使用 CUDA

# 設定你想要怎樣規格的虛擬機器.
rider@cloud:~$ sudo vim /etc/xen-tools/xen-tools.conf

dir = /home
install-method = debootstrap
size   = 8Gb      # Disk image size.
memory = 1024Mb    # Memory size
swap   = 128Mb    # Swap size
fs     = ext3     # use the EXT3 filesystem for the disk image.
dist   = hardy    # Default distribution to install. ---> For CUDA Support (Ubuntu 8.0.4)
image  = sparse   # Specify sparse vs. full disk images.
gateway   = 140.XXX.XXX.XXX
netmask   = 255.255.255.0
broadcast = 140.XXX.XXX.XXX
kernel      = /boot/vmlinuz-`uname -r`
initrd      = /boot/initrd.img-`uname -r`
mirror = http://gb.archive.ubuntu.com/ubuntu/
ext3_options   = noatime,nodiratime,errors=remount-ro
ext2_options   = noatime,nodiratime,errors=remount-ro
xfs_options    = defaults
reiser_options = defaults

rider@cloud:~$ sudo xen-create-image --hostname nvidia --ip 140.XXX.XXX.XXX

步驟三: 查看你的顯卡資訊

rider@cloud:~$ lspci -vv

01:00.0 VGA compatible controller: nVidia Corporation GeForce 9800 GT (rev a2)
	Subsystem: ASUSTeK Computer Inc. Device 82a0
	Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
	Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
	Latency: 0
	Interrupt: pin A routed to IRQ 16
	Region 0: Memory at fd000000 (32-bit, non-prefetchable) [size=16M]
	Region 1: Memory at d0000000 (64-bit, prefetchable) [size=256M]
	Region 3: Memory at fa000000 (64-bit, non-prefetchable) [size=32M]
	Region 5: I/O ports at dc80 [size=128]
	[virtual] Expansion ROM at fea00000 [disabled] [size=128K]
	Capabilities: <access denied>
	Kernel driver in use: pciback
	Kernel modules: nvidia, nvidiafb

步驟四: PCI Frontend Configuration 設定你的 DomU

rider@cloud:~$ sudo vim /etc/xen/nvidia.cfg

kernel      = '/boot/vmlinuz-2.6.22.9'
ramdisk     = '/boot/initrd.img-2.6.22.9'
 
memory      = '1024'
vcpus       = '4'

# 配置你的 PCIE 顯示卡
pci         = ['01:00.0']

root        = '/dev/sda2 ro'
disk        = [
                  'file:/home/domains/nvidia/disk.img,sda2,w',
                  'file:/home/domains/nvidia/swap.img,sda1,w',
              ]
name        = 'nvidia'
 
#
#  Networking
#
vif         = [ 'ip=140.xxx.xxx.xxx,mac=00:16:3E:AA:70:5C' ]
 
#
#  Behaviour
#
on_poweroff = 'destroy'
on_reboot   = 'restart'
on_crash    = 'restart'

步驟五: PCI Backend Configuration 設定你的 Dom0

rider@cloud:~$ sudo su -
# Hide the device from dom0 so pciback can take control.
root@cloud:~$ echo -n "0000:01:00.0" > /sys/bus/pci/drivers/nvidia/unbind

# Give the dev_ids to pciback, and give it a new slot then bind.
root@cloud:~$ echo -n "0000:01:00.0" > /sys/bus/pci/drivers/pciback/new_slot
root@cloud:~$ echo -n "0000:01:00.0" > /sys/bus/pci/drivers/pciback/bind

root@cloud:~$ cat /sys/bus/pci/drivers/pciback/slots

0000:01:00.0

# Caution: Make sure that the device is not controlled by any driver: there should be no driver symlink for nvidia.

PATH: /sys/bus/pci/devices/0000:01:00.0/
driver -> ../../../../bus/pci/drivers/nvidia ---> This symlink shouldn't exist.

步驟六: 硬體直接存取設定

Permissive Flag

rider@cloud:~$ sudo vim /etc/xen/xend-pci-permissive.sxp

(unconstrained_dev_ids
     #('0123:4567:89AB:CDEF')
     ('0000:01:00.0')
)

User-space Quirks

rider@cloud:~$ sudo vim /etc/xen/xend-pci-quirks.sxp

(pci_ids
   # Entries are formated as follows:  
   #     <vendor>:<device>[:<subvendor>:<subdevice>]

   ('10de:0605'   # NVIDIA 9800GT
   )
)

步驟七: 啟動並登入你的虛擬機器 DomU

說明: 用 root 免密碼先登入,然後建立自己的帳號. 改用自己的帳號登入(亦可用 root 登入 , 不新建帳號 ):
@ Dom0
rider@cloud:~$ sudo xm create -c nvidia.cfg

@ DomU
root@nvidia:~# adduser username
root@nvidia:~# vim /etc/sudoers

username    ALL=(ALL) ALL

步驟八: 設定你的 DomU 基本環境

# 設定 locales (系統語系)
rider@nvidia:~$ sudo vim /etc/profile

# Locale
export LANGUAGE="en_US.UTF-8"
export LC_ALL="en_US.UTF-8"
export LANG="en_US.UTF-8"

rider@nvidia:~$ source /etc/profile
rider@nvidia:~$ sudo dpkg-reconfigure locales
# 更新 PCI ID Database
rider@nvidia:~$ sudo apt-get update
rider@nvidia:~$ sudo apt-get install wget
rider@nvidia:~$ sudo update-pciids
# 查看顯卡資訊有無正常顯示
rider@nvidia:~$ lspci

00:00.0 VGA compatible controller: nVidia Corporation GeForce 9800 GT (rev a2)

# 查看顯卡資源有無順利分配到 DomU
rider@nvidia:~$ dmesg | grep pci

pcifront pci-0: Installing PCI frontend
pcifront pci-0: Creating PCI Frontend Bus 0000:00
pciback 0000:00:00.0: probing...
pciback: pcistub_init_devices_late

實作二: 在虛擬機器( Dom0 / DomU )上試跑 CUDA Examples

步驟九: 安裝 CUDA Toolkit & SDK

# 安裝環境所需套件

~$ sudo apt-get install autoconf automake build-essential gcc make mesa-common-dev libglu1-mesa-dev mesa-utils libxmu-headers libxmu6 libxmu-dev zlib1g-dev libjpeg62 libjpeg62-dev xutils-dev libxaw-headers libxaw7 libxaw7-dev libxext6 libxext-dev rxvt lwm xauth xvfb xfonts-100dpi xfonts-75dpi culmus xfonts-scalable xfonts-base libtool initramfs-tools libxi6 libxi-dev linux-kernel-devel xserver-xorg xserver-xorg-core xserver-xorg-dev

# 下載 NVIDIA CUDA toolkit
rider@nvidia:~$ mkdir -p ./nvidia/cuda
rider@nvidia:~$ cd ./nvidia/cuda/
rider@nvidia:~/nvidia/cuda$ wget http://developer.download.nvidia.com/compute/cuda/2_1/toolkit/cudatoolkit_2.1_linux64_ubuntu8.04.run

# 下載 NVIDIA CUDA SDK
rider@nvidia:~/nvidia/cuda$ wget http://developer.download.nvidia.com/compute/cuda/2_1/SDK/cuda-sdk-linux-2.10.1215.2015-3233425.run
rider@nvidia:~/nvidia/cuda$ chmod a+x *
rock@cloud:~$ sudo apt-get install autoconf automake build-essential gcc make libtool initramfs-tools libxi6 libxi-dev libxmu6 libxmu-dev xserver-xorg-core xserver-xorg-dev

# 安裝 NVIDIA CUDA toolkit (CUDA Compiler)
rider@nvidia:~/nvidia/cuda$ sudo sh cudatoolkit_2.1_linux64_ubuntu8.04.run

Enter install path (default /usr/local/cuda, '/cuda' will be appended):

# Note:

* Please make sure your PATH includes /usr/local/cuda/bin
* Please make sure your LD_LIBRARY_PATH includes /usr/local/cuda/lib
*   or add /usr/local/cuda/lib to /etc/ld.so.conf and run ldconfig as root

* Please read the release notes in /usr/local/cuda/doc/

* To uninstall CUDA, delete /usr/local/cuda
* Installation Complete

# 安裝 NVIDIA CUDA SDK (CUDA Runtime Libraries)
rock@cloud:~/nvidia/cuda$ sudo sh cuda-sdk-linux-2.10.1215.2015-3233425.run

# Note:

Enter install path (default /usr/local/cuda, '/cuda' will be appended): /usr/local/NVIDIA_CUDA_SDK

Configuring SDK Makefile (/usr/local/NVIDIA_CUDA_SDK/common/common.mk)...

* Please make sure your PATH includes /usr/local/cuda/bin
* Please make sure your LD_LIBRARY_PATH includes /usr/local/cuda/lib

* To uninstall the NVIDIA CUDA SDK, please delete /usr/local/NVIDIA_CUDA_SDK

# 設定 CUDA 執行環境
rider@cloud:~$ sudo su
root@cloud:~$ echo "export PATH=$PATH:/usr/local/cuda/bin" >> /etc/profile
root@cloud:~$ echo "export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib" >> /etc/profile
root@cloud:~$ source /etc/profile
root@cloud:~$ echo "/usr/local/cuda/lib" >> /etc/ld.so.conf
root@cloud:~$ ldconfig
root@cloud:~$ exit

步驟十: 試跑 CUDA 範例

# 換成 gcc-4.1 來編譯
rider@nvidia:~$ sudo apt-get install gcc-4.1 g++-4.1
rider@nvidia:~$ sudo rm /usr/bin/gcc
rider@nvidia:~$ sudo ln -sf /usr/bin/gcc-4.1 /usr/bin/gcc
# Linking library
rider@nvidia:~$ sudo ln -sf /usr/lib/libcuda.so.1 /usr/lib/libcuda.so
rider@nvidia:~$ sudo ln -sf /usr/lib/libglut.so.3 /usr/lib/libglut.so
# 進入 CUDA 專案目錄
rider@nvidia:~$ cd /usr/local/NVIDIA_CUDA_SDK/
# 編譯建構全部範例
rider@nvidia:/usr/local/NVIDIA_CUDA_SDK/$ sudo make
# 選擇一各 CUDA 範例
rider@nvidia:/usr/local/NVIDIA_CUDA_SDK/projects$ cd ./deviceQuery/
rider@nvidia:/usr/local/NVIDIA_CUDA_SDK/projects/deviceQuery$ sudo make
# 進入編譯完成的專案目錄
rider@nvidia:/usr/local/NVIDIA_CUDA_SDK/projects/deviceQuery$ cd ../../bin/linux/release/
# 執行
rider@nvidia:/usr/local/NVIDIA_CUDA_SDK/bin/linux/release$ sudo ./deviceQuery
# 輸出結果

There is 1 device supporting CUDA

Device 0: "GeForce 9800 GT"
  Major revision number:                         1
  Minor revision number:                         1
  Total amount of global memory:                 1073414144 bytes
  Number of multiprocessors:                     14
  Number of cores:                               112
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       16384 bytes
  Total number of registers available per block: 8192
  Warp size:                                     32
  Maximum number of threads per block:           512
  Maximum sizes of each dimension of a block:    512 x 512 x 64
  Maximum sizes of each dimension of a grid:     65535 x 65535 x 1
  Maximum memory pitch:                          262144 bytes
  Texture alignment:                             256 bytes
  Clock rate:                                    1.51 GHz
  Concurrent copy and execution:                 Yes

Test PASSED

Press ENTER to exit...