Database Reference
In-Depth Information
DeflateCompressor
The DeflateCompressor, more commonly known as Java's version of zip, is more
often used when you want better compression ratios. Your reads might come
a little slower, but your data on disk will be smaller than if you were using
SnappyCompressor.
File System
There are a number of options for which file system to use. If you are interested in
getting better use of your disk while still keeping some speed, it might be worth-
while to look into using ZFS. The common deployment file system for Linux is
ext4 and is already supported by most modern Linux distributions. Formatting a
device for ext4 is straightforward (see
Listing 6.1
)
, and the utility for doing so is
available out of the box.
Listing 6.1
Format a Drive with the ext4 File System
$ mke2fs -t ext4 /dev/md0
By default, ext4 is a fairly performant file system. By changing a few settings
(also known as mount options) in the /etc/fstab, you can get a little extra perform-
ance push.
noatime.
This stands for “no access time” updates. It tells the operating
system not to update the inode information on each access. This is useful
because Cassandra doesn't access inode information and there is no need
to pay the performance penalty for updating it.
barriers=0.
This option, when set to 0, disables write barriers. Write
barriers enforce the proper on-disk order of journal commits. Since ext4 is
a journal-based file system, you incur a performance penalty when ensur-
ing the order of the journals. With hardware-based systems, it is safe to
disable barriers if you have battery-backed disks (as is common to hard-
ware RAIDs).
data=journal, commit=15.
By default, the operating system will
sync all data (including metadata) to disk every five seconds. By extend-
ing this value, you can improve performance.
commit
applies only if you
set
data=journal
.
