在虚拟机上部署SPDK NVME测试案例

 作者简介:

           万群

Intel 存储软件工程师

主要从事SPDK软件测试工作

640?wx_fmt=gif_1 背景介绍_

正如我们之前在微信中所讨论的在物理机器[1]上使用SoftRoCE部署SPDK NVMe-oF的方法一样,现在,我们将要讨论在虚拟机上使用部署SPDK NVMe测试案例[2]的方法。为什么我们需要在虚拟机上部署它呢?原因很明显随着主机CPU Core越来越强大,我们需要有效地利用CPU Core,而对于不同种类且不需要太多计算和内存资源的功能测试来说,我们可以充分利用虚拟机运行功能测试如NVMevhost i SCSI测试。限于篇幅,这篇文章主要介绍如何在VM上部署SPDK NVME测试案例。后面会有陆续介绍vhost iS CSI的介绍。同时在9月北京的峰会我们也会做一些相应的介绍。
希望通过这篇文章能让大家熟悉SPDK,不管在真实NVMe SSD场景下,还是在虚拟环境下。

640?wx_fmt=gif

_2 配置虚拟机环境_

在Windows10(1607)操作系统的机器上配置虚拟机的步骤如下所示:

1.安装 VirtualBox (https://www.virtualbox.org/wiki/Downloads 及其相关组件,需要安装两个安装包
  • 在BIOS里面确保VT选项是打开的,安装VirtualBox5.1.38 或者更新版本(建使用5.1.38版本)

  • 搜索“Windows features”并关闭Hyper-V选项,重启windows 操作系统,安VirtualBox ExtensionPack(https://www.virtualbox.org/wiki/Downloads5.1.38版本或者更新版本(同样建议使用5.1.38版本)

2.安装Vagrant(https://www.vagrantup.com/)1.9.4版本或者更新版本(建议使用1.9.4版本)

3.安装 Git Bash (https://www.git-scm.com/download/win) (版本不做要求)

4.打开Gitbash终端,下载SPDK代码并创建虚拟机环境

a.如果有防火墙,建议使用代理来提高下载速度

  i.设置http_proxy和https_proxy代理,命令行如下
$export https_proxy=
$export http_proxy=
  ii.安装proxyconf plugin,命令如下
$vagrant plugin install vagrant-proxyconf

b.下载SPDK代码,命令行如下

$git clone https://github.com/spdk/spdk.git

c.编译并创建虚拟机,命令行如下

$cd spdk/
$git submodule update –init
$./spdk/scripts/vagrant/create_vbox.sh –n 4  –s 4096 fedora28
-n指定虚拟机CPU数量,-s指定虚拟机内存大小,执行这一步如遇到下面的问题,则很有可能是网络不够稳定或者网速太慢,可以选择一个网速比较稳定或者快的时间重新下载

d.登录虚拟机,命令如下

$cd fedora28-virtualbox
$vagrant ssh
恭喜你!从现在开始,你可以登录虚拟机工作了

e.验证虚拟NVMe设备是否存在,命令行如下

$lspci | grep “Non-Volatile”
00:0e.0 Non-Volatile  memory controller: InnoTek Systemberatung GmbH Device 4e56

f.同样,在虚拟机里面,也需要对代理进行设置,然后编译SPDK代码,命令行如下:

$apt-get update
$cd ~/spdk_repo/spdk
$git submodule update –init
如果无法成功执行上面的命令,可以尝试重新下载SPDK代码,命令为
$git  clone https://github.com/spdk/spdk.git
$scripts/pkgdep.sh
如果出现警告可以使用如下命令进行清除
$git reset --hard
$make

g.通过运行hello_word实例来验证SPDK安装成功

$script/setup.sh
$cd examples/bdev/hello_world/
$./hello_bdev –c bdev.conf

640?wx_fmt=gif
_3 在VM上部署SPDK NVMe测试案例_

examples目录运行NVMe测试案例


1.运行setup.sh脚本
$cd /home/vagrant/spdk_repo/spdk
$./scripts/setup.sh
0000:00:0e.0 (80ee 4e56): nvme -> uio_pci_generic
2.在examples目录下运行NVMehello_world测试案例来确保环境已经能够正常使用虚拟的nvme设备
$ ./examples/nvme/hello_world/hello_world
Starting SPDK v19.10-pre / DPDK 19.05.0 initialization...
[ DPDK EAL parameters: hello_world -c 0x1  --log-level=lib.eal:6 --log-level=lib.cryptodev:5 --log-level=user1:6  --base-virtaddr=0x200000000000 --match-allocations --file-prefix=spdk0  --proc-type=auto ]
Initializing NVMe Controllers
Attaching to 0000:00:0e.0
Attached to 0000:00:0e.0
Using controller ORCL-VBOX-NVME-VER12 (VB1234-56789        ) with 1 namespaces.
  Namespace ID: 1  size: 1GB
Initialization complete.
INFO: using host memory buffer for IO
Hello world!
3.在examples目录下运行NVMeperf测试案例测试各种IO的性能
$./examples/nvme/perf/perf  -q 128 -o 4096 -w randread -r 'trtype:PCIe traddr:0000:00:0e.0' -t 10
Starting SPDK v19.10-pre / DPDK 19.05.0 initialization...
[ DPDK EAL parameters: perf --no-shconf -c 0x1  --log-level=lib.eal:6 --log-level=lib.cryptodev:5 --log-level=user1:6  --base-virtaddr=0x200000000000 --match-allocations --file-prefix=spdk_pid7543  ]
Initializing NVMe Controllers
Attaching to NVMe Controller at 0000:00:0e.0
Attached to NVMe Controller at 0000:00:0e.0 [80ee:4e56]
Associating PCIE (0000:00:0e.0) with lcore 0
Initialization complete. Launching workers.
Starting thread on core 0
========================================================
                                                                     Latency(us)
Device Information             :       IOPS      MiB/s     Average        min        max
PCIE (0000:00:0e.0) from core 0:    6257.30      24.44    20476.55    2274.99   31814.34
========================================================
Total                          :    6257.30      24.44    20476.55    2274.99   31814.34
4.在examples目录下运行NVMenvme_manage测试案例,这里nvme_manage可用来查看controller的信息,创建namespace以及更新firmware等,具体可以通过选择相应数字选项来操作
$./examples/nvme/nvme_manage/nvme_manage
Starting SPDK v19.10-pre / DPDK 19.05.0 initialization...
[ DPDK EAL parameters: nvme_manage --no-shconf -c 0x1  --log-level=lib.eal:6 --log-level=lib.cryptodev:5 --log-level=user1:6 --base-virtaddr=0x200000000000  --match-allocations --file-prefix=spdk_pid7547 ]
NVMe Management Options
        [1: list  controllers]
        [2: create  namespace]
        [3: delete  namespace]
        [4: attach  namespace to controller]
        [5: detach  namespace from controller]
        [6: format  namespace or controller]
        [7: firmware  update]
        [8: opal]
        [9: quit]
5.在examples目录下运行NVMeidentify测试案例, 用于查看SSDvendorIDfirmware以及型号等信息
$ ./examples/nvme/identify/identify
Starting SPDK v19.10-pre / DPDK 19.05.0 initialization...
[ DPDK EAL parameters: identify --no-shconf -c 0x1 -n 1 -m  0 --log-level=lib.eal:6 --log-level=lib.cryptodev:5 --log-level=user1:6  --base-virtaddr=0x200000000000 --match-allocations --file-prefix=spdk_pid7554  ]
=====================================================
NVMe Controller at 0000:00:0e.0 [80ee:4e56]
=====================================================
Controller Capabilities/Features
================================
Vendor ID:                             80ee
Subsystem Vendor ID:                   80ee
Serial Number:                         VB1234-56789
Model Number:                           ORCL-VBOX-NVME-VER12
Firmware Version:                      1.0
Recommended Arb Burst:                 0
IEEE OUI Identifier:                   00 00 00
Multi-path I/O
  May have multiple  subsystem ports:   No
  May be connected to  multiple hosts:  No
  Associated with  SR-IOV VF:           No
Max Data Transfer Size:                Unlimited
Error Recovery Timeout:                 Unlimited
NVMe Specification Version (VS):       1.2
NVMe Specification Version (Identify): 1.2
Maximum Queue Entries:                 4096
Contiguous Queues Required:            Yes
Arbitration Mechanisms Supported
  Weighted Round  Robin:                Not Supported
  Vendor  Specific:                     Not  Supported
Reset Timeout:                         5000 ms
Doorbell Stride:                       4 bytes
NVM Subsystem Reset:                   Not Supported
Command Sets Supported
  NVM Command  Set:                     Supported
Boot Partition:                        Not Supported
Memory Page Size Minimum:              4096 bytes
Memory Page Size Maximum:              4096 bytes
Optional Asynchronous Events Supported
  Namespace Attribute  Notices:         Not Supported
  Firmware Activation  Notices:         Not Supported
128-bit Host Identifier:               Not Supported
限于篇幅,后续的打印日志省略
 

test目录运行nvme测试案例


1.运行aer测试案例
$ ./test/nvme/aer/aer -r 'trtype:PCIe traddr:0000:00:0e.0'
Starting SPDK v19.10-pre / DPDK 19.05.0 initialization...
[ DPDK EAL parameters: aer --no-shconf -c 0x1  --log-level=lib.eal:6 --log-level=lib.cryptodev:5 --log-level=user1:6  --base-virtaddr=0x200000000000 --match-allocations --file-prefix=spdk_pid7715  ]
Asynchronous Event Request test
Attaching to 0000:00:0e.0
Attached to 0000:00:0e.0
Registering asynchronous event callbacks...
Cleaning up...
2.运行e2edp测试案例
$ ./test/nvme/e2edp/nvme_dp
Starting SPDK v19.10-pre / DPDK 19.05.0 initialization...
[ DPDK EAL parameters: nvme_dp -c 0x1 --log-level=lib.eal:6  --log-level=lib.cryptodev:5 --log-level=user1:6  --base-virtaddr=0x200000000000 --match-allocations --file-prefix=spdk0  --proc-type=auto ]
NVMe Write/Read with End-to-End data protection test
Attaching to 0000:00:0e.0
Attached to 0000:00:0e.0
Cleaning up...
3.运行run_perf.sh命令
$ ./test/nvme/perf/run_perf.sh --numjobs=1 --iodepth=128  --rw=randread --fio-bin=/usr/src/fio/fio --ramp-time=10 --run-time=10
++++ dirname ./test/nvme/perf/run_perf.sh
+++ readlink -f ./test/nvme/perf
++ BASE_DIR=/home/vagrant/spdk/test/nvme/perf
+++ readlink -f /home/vagrant/spdk/test/nvme/perf/../../..
++ ROOT_DIR=/home/vagrant/spdk
++  PLUGIN_DIR_NVME=/home/vagrant/spdk/examples/nvme/fio_plugin
++  PLUGIN_DIR_BDEV=/home/vagrant/spdk/examples/bdev/fio_plugin
++ BDEVPERF_DIR=/home/vagrant/spdk/test/bdev/bdevperf
++ . /home/vagrant/spdk/scripts/common.sh
++ . /home/vagrant/spdk/test/common/autotest_common.sh
+++ xtrace_disable
+++ PREV_BASH_OPTS=ehxB
+++ set +x
   01:27:46     # xtrace_enable
  01:27:46      #  NVME_FIO_RESULTS=/home/vagrant/spdk/test/nvme/perf/result.json
  01:27:46      # PRECONDITIONING=true
  01:27:46      # FIO_BIN=/usr/src/fio/fio
  01:27:46      # RUNTIME=600
  01:27:46      # PLUGIN=nvme
  01:27:46      # RAMP_TIME=30
  01:27:46      # BLK_SIZE=4096
  01:27:46      # RW=randrw
  01:27:46      # MIX=100
  01:27:46      # IODEPTH=256
  01:27:46      # DISKNO=1
  01:27:46      # ONEWORKLOAD=false
  01:27:46      # CPUS_ALLOWED=1
  01:27:46      # NUMJOBS=1
  01:27:46      # REPEAT_NO=3
  01:27:46      # NOIOSCALING=false
  01:27:46      # getopts h-: optchar
  01:27:46      # case "$optchar" in
  01:27:46      # case "$OPTARG" in
  01:27:46      # NUMJOBS=1
限于篇幅,这里只列出前面几行的打印日志
最后的测试性能数据保存在下列文件中 : /home/vagrant/spdk/test/nvme/perf/results/perf_results_4096BS_128QD_randread_100MIX_nvme_08_05_2019_012746/perf_results_4096BS_128QD_randread_100MIX_nvme_08_05_2019_012746.csv

640?wx_fmt=gif

_4 问题解答_

1.在虚拟机中使用自动测试运行时,如果遇到以下问题,说明虚拟机没有给虚拟机中的大页面分配足够的内存,则需要在虚拟机创建时分配更多内存,如 12288KB
## ERROR: requested 4096 hugepages but only  1719 could be allocated.
## Memory might be heavily fragmented. Please try flushing the system  cache, or reboot the machine.
0000:00:0e.0 (80ee 4e56): uio_pci_generic -> nvme
Clean
2. 当用下列命令创建虚拟机的时候,如果遇到网速不稳定或者网速较慢的时候,会遇到如下问题,可以通过选择一个网速较快的时段下
$/spdk/scripts/vagrant/create_vbox.sh –n 4 –s4096 fedora28
Errormessage: An error occurred while downloading the remote file. The error
message,if any, is reproduced below. Please fix this error and try again. OpenSSLSSL_read: SSL_ERROR_SYSCALL, errno 10054

参考文献

[1] https://mp.weixin.qq.com/s/--06WT3NCVxDXVEbD7txNA
[2] https://dqtibwqq6s6ux.cloudfront.net/


Thanks for watching!

END

640?wx_fmt=gif

    转载须知   

 推荐阅读 

SPDK NVMe Reservation使用简介

在虚拟机上部署SPDK NVME测试案例_第1张图片

你可能感兴趣的:(在虚拟机上部署SPDK NVME测试案例)