Vertica Analytics Platform Version 9.2.x Documentation
This topic details how to change I/O Scheduling to a supported scheduler. Vertica requires that I/O Scheduling be set to
noop. The installer checks what scheduler the system is using, reporting an unsupported scheduler issue with identifier: S0150. If the installer cannot detect the type of scheduler in use (typically if your system is using a RAID array), it reports that issue with identifier: S0151.
If your system is not using a RAID array, then complete the following steps to change your system to a supported I/O Scheduler. If you are using a RAID array, then consult your RAID vendor documentation for the best performing scheduler for your hardware.
Configure the I/O Scheduler
The Linux kernel can use several different I/O schedulers to prioritize disk input and output. Most Linux distributions use the Completely Fair Queuing (CFQ) scheme by default, which gives input and output requests equal priority. This scheduler is efficient on systems running multiple tasks that need equal access to I/O resources. However, it can create a bottleneck when used on Vertica drives containing the catalog and data directories, because it gives write requests equal priority to read requests, and its per-process I/O queues can penalize processes making more requests than other processes.
Instead of the CFQ scheduler, configure your hosts to use either the Deadline or NOOP I/O scheduler for the drives containing the catalog and data directories:
- The Deadline scheduler gives priority to read requests over write requests. It also imposes a deadline on all requests. After reaching the deadline, such requests gain priority over all other requests. This scheduling method helps prevent processes from becoming starved for I/O access. The Deadline scheduler is best used on physical media drives (disks using spinning platters), since it attempts to group requests for adjacent sectors on a disk, lowering the time the drive spends seeking.
- The NOOP scheduler uses a simple FIFO approach, placing all input and output requests into a single queue. This scheduler is best used on solid state drives (SSDs). Because SSDs do not have a physical read head, no performance penalty exists when accessing non-adjacent sectors.
Failure to use one of these schedulers for the Vertica drives containing the catalog and data directories can result in slower database performance. Other drives on the system (such as the drive containing swap space, log files, or the Linux system files) can still use the default CFQ scheduler (although you should always use the NOOP scheduler for SSDs).
There are two ways for you to set the scheduler used by your disk devices:
Write the name of the scheduler to a file in the
- Use a kernel boot parameter.
Configure the I/O Scheduler - Changing the Scheduler Through the /sys Directory
You can view and change the scheduler Linux uses for I/O requests to a single drive using a virtual file under the
/sys directory. The name of the file that controls the scheduler a block device uses is:
Where deviceName is the name of the disk device, such as
cciss\!c0d1 (the first disk on an Micro Focus RAID array). Viewing the contents of this file shows you all of the possible settings for the scheduler. The currently-selected scheduler is surrounded by square brackets:
# cat /sys/block/sda/queue/scheduler noop deadline [cfq]
To change the scheduler, write the name of the scheduler you want the device to use to its scheduler file. You must have root privileges to write to this file. For example, to set the sda drive to use the deadline scheduler, run the following command as root:
# echo deadline > /sys/block/sda/queue/scheduler # cat /sys/block/sda/queue/scheduler noop [deadline] cfq
Changing the scheduler immediately affects the I/O requests for the device. The Linux kernel starts using the new scheduler for all of the drive's input and output requests.
While tests show that changing the scheduler settings while Vertica is running does not cause problems, Vertica recommends shutting down. Before changing the I/O schedule, or making any other changes to the system configuration, consider shutting down any running database.
Changes to the I/O scheduler made through the
/sys directory only last until the system is rebooted, so you need to add the commands that change the I/O scheduler to a startup script (such as those stored in
/etc/init.d, or though a command in
/etc/rc.local). You also need to use a separate command for each drive on the system whose scheduler you want to change.
For example, to make the configuration take effect immediately and add it to rc.local so it is used on subsequent reboots.
For systems that do not support
/etc/rc.local, use the equivalent startup script that is run after the destination runlevel has been reached. For example SuSE uses
echo deadline > /sys/block/sda/queue/scheduler echo 'echo deadline > /sys/block/sda/queue/scheduler' >> /etc/rc.local
On some Ubuntu/Debian systems, the last line in rc.local must be "
exit 0". So you must manually add the following line to
etc/rc.local before the last line with
You may prefer to use this method of setting the I/O scheduler over using a boot parameter if your system has a mix of solid-state and physical media drives, or has many drives that do not store Vertica catalog and data directories.
If you are using Red Hat 7.0 or CentOS 7.0 or higher, run the following command as root or sudo:
$ chmod +x /etc/rc.d/rc.local
Configure the I/O Scheduler - Changing the Scheduler with a Boot Parameter
elevator kernel boot parameter to change the default scheduler used by all disks on your system. This is the best method to use if most or all of the drives on your hosts are of the same type (physical media or SSD) and will contain catalog or data files. You can also use the boot parameter to change the default to the scheduler the majority of the drives on the system need, then use the
/sys files to change individual drives to another I/O scheduler. The format of the elevator boot parameter is:
Where schedulerName is
cfq. You set the boot parameter using your bootloader (grub or grub2 on most recent Linux distributions). See your distribution's documentation for details on how to add a kernel boot parameter.
Was this topic helpful?