什么是SSD TRIM (by quqi99)

作者：张华  发表于：2016-03-23

版权声明：可以任意转载，转载时请务必以超链接形式标明文章原始出处和作者信息及本版权声明

( http://blog.csdn.net/quqi99 )

 

普通硬盘并不真正从硬盘里删除数据，只是在操作系统标记为删除，下次直接覆盖写；但SSD物理覆盖写却是很慢的(对于SSD，可以以4KB大小的页(128个page组成一个block)来读写数据，但却只能以512KB(128 page)大小来删除。要覆盖写一个4KB page的话，先要把512KB block全部复制到缓存里，然后删除这4kb page并替换成新的写数据，接下来清空SSD的这512KB block区域，并从缓存里把新的数据写回去）。所以当SSD的所有空闲块都被使用后，再有写的操作只能覆盖写到之前被操作系统标记为删除的区域，这也是速度下降的开始。特别是一系列小文件随机写入同时覆盖写很多的块，那样缓存一下子会快速过载而变得非常慢，所以SSD厂商都在新主控制器上加入了越来越大的外置缓存，这虽然可以一定程序上解决随机写入卡的问题，但是却不能解决SSD在覆盖写时速度下降的问题。如果操作系统在删除一个4KB page的时候不仅在操作系统记录里删除也控制SSD主控物理删除它的话，这就是TRIM干的事情，它将写时的延迟移到了删除时。需要内核、固件（SSD厂商在固件里放TRim算法）、驱动（如Intel的AHCI驱动）三者都支持TRIM才行。

可用fstrim命令回收一个已挂载的文件系统上所有未使用的块（sudo fstrim -v）；

也可使用smartctl -a /dev/sda 命令检测磁盘是否有错，从结果里找类似"177 Wear_Leveling_Count 0x0013 099 099 000 Pre-fail Always - 9"，那个9是SSD厂商估计的寿命，0代表新SSD盘，100代表可能明天就坏了。

 

下面是smartmontools自检工具的基本使用

1, 安装包

   sudo apt-get install smartmontools

2, 查看硬盘是否enable了SMART自检特性，若没enable,可使用命令”sudo smartctl --smart=on --offlineauto=on --saveauto=on /dev/sda“打开。
hua@node1:~$ sudo smartctl -i /dev/sda
smartctl 6.2 2013-07-26 r3841 [x86_64-linux-4.4.0-9-generic] (local build)
Copyright (C) 2002-13, Bruce Allen, Christian Franke, www.smartmontools.org


=== START OF INFORMATION SECTION ===
Model Family:     Seagate Barracuda 7200.14 (AF)
Device Model:     ST2000DM001-9YN164
Serial Number:    W1E0S44Q
LU WWN Device Id: 5 000c50 052bb6b91
Firmware Version: CC4B
User Capacity:    2,000,398,934,016 bytes [2.00 TB]
Sector Sizes:     512 bytes logical, 4096 bytes physical
Rotation Rate:    7200 rpm
Device is:        In smartctl database [for details use: -P show]
ATA Version is:   ATA8-ACS T13/1699-D revision 4
SATA Version is:  SATA 3.0, 6.0 Gb/s (current: 6.0 Gb/s)
Local Time is:    Wed Mar 23 14:50:55 2016 CST


==> WARNING: A firmware update for this drive may be available,
see the following Seagate web pages:
http://knowledge.seagate.com/articles/en_US/FAQ/207931en
http://knowledge.seagate.com/articles/en_US/FAQ/223651en


SMART support is: Available - device has SMART capability.
SMART support is: Enabled

3, 检测
hua@node1:~$ sudo smartctl -H /dev/sda
smartctl 6.2 2013-07-26 r3841 [x86_64-linux-4.4.0-9-generic] (local build)
Copyright (C) 2002-13, Bruce Allen, Christian Franke, www.smartmontools.org


=== START OF READ SMART DATA SECTION ===
SMART overall-health self-assessment test result: PASSED


4, 详细检测
hua@node1:~$ sudo smartctl -A /dev/sda
smartctl 6.2 2013-07-26 r3841 [x86_64-linux-4.4.0-9-generic] (local build)
Copyright (C) 2002-13, Bruce Allen, Christian Franke, www.smartmontools.org


=== START OF READ SMART DATA SECTION ===
SMART Attributes Data Structure revision number: 10
Vendor Specific SMART Attributes with Thresholds:
ID# ATTRIBUTE_NAME          FLAG     VALUE WORST THRESH TYPE      UPDATED  WHEN_FAILED RAW_VALUE
  1 Raw_Read_Error_Rate     0x000f   117   099   006    Pre-fail  Always       -       141775376
  3 Spin_Up_Time            0x0003   095   094   000    Pre-fail  Always       -       0
  4 Start_Stop_Count        0x0032   100   100   020    Old_age   Always       -       561
  5 Reallocated_Sector_Ct   0x0033   100   100   036    Pre-fail  Always       -       0
  7 Seek_Error_Rate         0x000f   084   060   030    Pre-fail  Always       -       306606505
  9 Power_On_Hours          0x0032   095   095   000    Old_age   Always       -       5237
 10 Spin_Retry_Count        0x0013   100   100   097    Pre-fail  Always       -       0
 12 Power_Cycle_Count       0x0032   100   100   020    Old_age   Always       -       667
183 Runtime_Bad_Block       0x0032   100   100   000    Old_age   Always       -       0
184 End-to-End_Error        0x0032   100   100   099    Old_age   Always       -       0
187 Reported_Uncorrect      0x0032   099   099   000    Old_age   Always       -       1
188 Command_Timeout         0x0032   100   099   000    Old_age   Always       -       0 0 1
189 High_Fly_Writes         0x003a   100   100   000    Old_age   Always       -       0
190 Airflow_Temperature_Cel 0x0022   067   054   045    Old_age   Always       -       33 (Min/Max 20/36)
191 G-Sense_Error_Rate      0x0032   100   100   000    Old_age   Always       -       0
192 Power-Off_Retract_Count 0x0032   100   100   000    Old_age   Always       -       77
193 Load_Cycle_Count        0x0032   100   100   000    Old_age   Always       -       1907
194 Temperature_Celsius     0x0022   033   046   000    Old_age   Always       -       33 (0 8 0 0 0)
197 Current_Pending_Sector  0x0012   100   100   000    Old_age   Always       -       0
198 Offline_Uncorrectable   0x0010   100   100   000    Old_age   Offline      -       0
199 UDMA_CRC_Error_Count    0x003e   200   200   000    Old_age   Always       -       0
240 Head_Flying_Hours       0x0000   100   253   000    Old_age   Offline      -       5111h+51m+54.467s
241 Total_LBAs_Written      0x0000   100   253   000    Old_age   Offline      -       12778926325684
242 Total_LBAs_Read         0x0000   100   253   000    Old_age   Offline      -       42981342412817

 

虚机中的discard

通过将trim与unmap指令提供的功能叫discard - search: DISCARD SUPPORT IN QEMU-KVM
对于SSD, trim指令允许操作系统在SSD上将不再使用的数据通知到SSD底层并在其内部将数据擦除掉。
虚机的vdisk与SSD类似, 也存在若不及时删除数据的话会造成数据越来越大, 这样virtio-scsi虚拟磁盘上出现了一个类似trim的指令叫unmap. 

1, 虚机定义增加额外的discard="unmap" (sda)
NOTE: 只对virtio-scsi驱动有效: glance image-update --property hw_scsi_model=virtio-scsi --property hw_disk_bus=scsi
   
     
     
     
   
   
     
     
     
   
2, 虚机内部挂载磁盘时增加对discard的支持: mount -o discard /dev/sda1 /sda
3, 测试脚本
$ for _ in {1..3}; do
    /bin/dd if=/dev/urandom of=/sda/file.$RANDOM bs=1M count=100 && sync
    /bin/dd if=/dev/urandom of=/sdb/file.$RANDOM bs=1M count=100 && sync
    /bin/rm -f /sda/file.* /sdb/file.* && sync
4, 使用fstrim命令手动一次性触发discard操作: /mntpoint. 或者如果使用上面脚试脚本后应该在host机器上用下列命令观察到自动trim
$ du -h sd{a,b}.img
34M  sda.img
138M sdb.img

另外, ceph驱动也支持trim, 除了使用virtio-scsi驱动(glance image-update --property hw_scsi_model=virtio-scsi --property hw_disk_bus=scsi), 也需要编辑nova.conf的[libvirt]段添加hw_disk_discard = unmap (注: cinder通过[ceph] section中report_discard_supported = True 来支持discard, so it is possible to configure Cinder per backend.), 这时虚机的libvirt xml配置文件应该有所反应, 或者 , 在虚机里面查看/sys/block/sdb/queue/discard_*, 见: 
https://ceph.io/geen-categorie/openstack-and-ceph-rbd-discard/
http://www.sebastien-han.fr/blog/2016/05/11/OpenStack-Cinder-discard-support-for-Ceph-in-Mitaka/

# $cinder/cinder/volume/manager.py
1759         # Add discard flag to connection_info if not set in the driver and
1760         # configured to be reported.
1761         if conn_info['data'].get('discard') is None:
1762             discard_supported = (self.driver.configuration
1763                                  .safe_get('report_discard_supported'))                                                                                                           
1764             if discard_supported:
1765                 conn_info['data']['discard'] = True

# $nova/nova/virt/libvirt/volume/volume.py
 93         # Configure usage of discard                                                                                                                                               
 94         if data.get('discard', False) is True:
 95             conf.driver_discard = 'unmap'

#./nova/virt/libvirt/imagebackend.py
 82 class Image(object):
  86     def __init__(self, path, source_type, driver_format, is_block_dev=False):
            ...
105         self.discard_mode = CONF.libvirt.hw_disk_discard 

 148     def libvirt_info(self, ...):
166         info.driver_discard = self.discard_mode

# $nova/nova/virt/libvirt/config.py
1009             if self.driver_discard is not None:
1010                 drv.set("discard", self.driver_discard)

# $nova/virt/libvirt/driver.py
 1765     def attach_volume(self, ...)
              ...
 1792         disk_info = blockinfo.get_info_from_bdm(
 1793             instance, CONF.libvirt.virt_type, instance.image_meta, bdm)
 1794         if disk_info['bus'] == 'scsi':
 1795             disk_info['unit'] = self._get_scsi_controller_max_unit(guest) + 1

 1797         conf = self._get_volume_config(connection_info, disk_info)
 1798 
 1799         self._check_discard_for_attach_volume(conf, instance)


 1748     def _check_discard_for_attach_volume(self, conf, instance):
 1749         """Perform some checks for volumes configured for discard support.
 1750 
 1751         If discard is configured for the volume, and the guest is using a
 1752         configuration known to not work, we will log a message explaining
 1753         the reason why.
 1754         """
 1755         if conf.driver_discard == 'unmap' and conf.target_bus == 'virtio':                                                                                                       
 1756             LOG.debug('Attempting to attach volume %(id)s with discard '
 1757                       'support enabled to an instance using an '
 1758                       'unsupported configuration. target_bus = '
 1759                       '%(bus)s. Trim commands will not be issued to '
 1760                       'the storage device.',
 1761                       {'bus': conf.target_bus,
 1762                        'id': conf.serial},
 1763                       instance=instance)