White Papers

Performance tuning

43 Dell EMC SC Series: Red Hat Enterprise Linux Best Practices | CML1031

unnecessarily high levels of I/O control traffic as well as raised CPU loads, causing an increase in I/O latency

and potentially impacting business-critical I/O.

An alternative to this implementation is using the fstrim command. The fstrim command is part of the util-

linux package and allows a one-time request to discard used blocks from a mounted filesystem. This

command can be scripted (shown in the following sample script), injected into a scheduled cron job, and then

applied to a set of mounted filesystems in batch during a time of day that would have less impact to business-

critical functions.

#!/bin/bash

FSTrim=/usr/bin/fstrim

MntPoints="u01 u02 u03 u04 u05"

for i in ${MntPoints}

echo "INFO: Applying ${FSTrim} to mount point ${i}"

${FSTrim} -v /${i}

done

4.5 SCSI device queue settings

Several Linux SCSI device queue settings can be applied to tune performance. The more common ones are

listed in the following sections with a brief explanation of what each parameter does with regard to I/O. These

values are found in the /sys/block/dm-X/queue directory (multipath devices) and /sys/block/sdX/queue

directory (block devices) and should be modified for each path device for the intended multipath volume.

4.5.1 I/O scheduler

The /sys/block/<device>/queue/schedule parameter and its contents define the I/O scheduler in use by the

Linux kernel for SCSI devices. Some application vendors (such as Oracle) provide specific recommendations

for which I/O scheduler to use in order to achieve optimal performance with the application platform. The

default scheduler varies on different versions of RHEL. Table 1 summaries the available schedulers and the

preferred schedulers to use with SC Series storage system. The scheduler in use is denoted by the [] bracket.

It is recommended to use deadline or mq-deadline scheduler for all SC Series array implementations. The

scheduler can be further configured by writing directly to the devices’ sysfs file (such as fifo_batch,

read_expire, and write_expire) that are discussed in the Performance Tuning Guide on the Red Hat

Enterprise Linux Document Portal.

Preferred I/O schedulers

RHEL 6 and older

RHEL 7

RHEL 8

noop anticipatory [deadline] cfq

noop [deadline] cfq

none [mq-deadline] kyber bfq

Note: RHEL 8 supports a new multi-queue deadline scheduling on block devices. Multi-queue deadline

scheduling allows better block layer performance with SSDs and multi-core systems.