Database Reference
In-Depth Information
the disks and prevents data loss during power outages; it is configured with an internal power backup mechanism.
Certain storage devices, for example, Oracle's ZFS Storage, perform the write to disk activity in parallel while the cache
is being populated, thus protecting the data during power outages.
There are two types of cache that can be managed at the disk/storage level: read cache and write cache.
Read Cache
Controllers attempt to “read ahead” (read-ahead policy), anticipating future read requests from the OS and buffering
from sequential sectors of the logical drive when seeking data. There are benefits of using the read-ahead policy, as
data is written sequentially. Read cache holds data that has been previously read, making subsequent reads of the
same data sets faster by eliminating the need for the controller to go back to the disk for the data.
Write Cache
Write cache can be implemented in one of two ways:
•
Write-Back.
With this option, the controller sends a write-request completion signal as soon as
the data is in the controller cache; it does not wait for the data to be written to disk. Write-back
caching may provide improved performance since subsequent read requests can retrieve data
from the controller cache more quickly than when reading it from the disk. The greater the size
of the cache, the more data is buffered, thus improving throughput and IOPS of data written.
•
Write-Through.
With this option, the controller sends a write-request completion signal only
after the data has been successfully written to disk. Write-through caching provides better
data security compared to write-back caching, since the system assumes the data is available
only after it has been safely written to the disk.
Apart from the configuration and sizing of parameters, performance of the I/O system also depends on the
following:
•
How the LUNs have been carved
•
What type of RAID has been configured
•
Whether the logical drives reside on the same array
•
How the data spread across all the members of the array, including the number of drives being
used as well as the size and speed of these devices or disks
All of these variables can have a great impact on the overall performance of the I/O subsystem.
ZFS DISK CaChe
ZFS (Zettabyte File System) guarantees that the data is written on disk before the write system call completes, even
if write cache is enabled. When a synchronized write i/o operation is performed, the write transaction will be written
to the ZFS intent log (ZiL) first. once the transaction is saved to the ZiL on the physical disk platter (not on write
cache), ZFS can replay the i/o, even if a sudden power outage happens before the data is updated to the file system.
therefore, even if write cache is enabled on disk, ZFS can guarantee the data is written by synchronized i/o in case
of a power outage. however, if synchronized i/o is not enabled, the data will be written to disk several (or more)
seconds after the write system call returns. in this case, an unexpected power failure may cause data loss.
5
5
MOS Note: 1122223.1 How ZFS Manages Write Cache on Disk.
Search WWH ::
Custom Search