Database Reference
In-Depth Information
Storage Tiering
Storage tiers have been common in the storage array for quite some time. Storage engineering maps different classes
of storage to tiers, often with the help of the storage array's vendor. The different classes of storage are taken from
a matrix of storage types such as DRAM, flash memory, spinning disk, and performance attributes such as RAID
levels. A third dimension to the matrix is the transport protocol. Current mainstream transport media include Fiber
Channel, Fiber Channel over Ethernet and iSCSI as well as Network File System. As you can imagine, the number of
permutations of these are large, and it requires careful planning for which combination should be made available as a
storage tier. One approach to storage tiering relying solely on hard disks could resemble the following:
1.
15k RPM Fiber Channel disks in a RAID 10
2.
15k RPM Fiber Channel disks in a RAID 5
3.
10k RPM Fiber Channel disks in RAID 5
4.
Direct NFS with dedicated appliances
In the above list, the lowest number should be the “best” storage tier. Storage tiering has often been used to
enable organizations to implement data life-cycle models. Frequently accessed, or “hot,” data was placed on the
better-quality storage tiers, and cold, or very infrequently used, data was placed on lower-end storage. In the Oracle
world, that often meant placing objects on tablespaces defined within a storage tier. Moving data from one tier to
another required either the use of “ alter table ... move ” commands or alternatively the implementation of calls to
the database package DBMS_REDEFINITION . Needless to say, this was a non-automated task that had to be performed
during maintenance windows. High-end enterprise arrays nowadays try to perform the same task automatically in the
background. Predictable performance is more important than stellar performance on day one, followed by abysmal
performance on day two. Time will tell if the automated models are sophisticated and stable enough to guarantee
consistent execution times and adherence to the agreed service levels.
The commoditization of flash storage has changed the classic storage tier model presented in the preceding list.
The introduction of flash memory, or what some call solid state disk , has fundamentally changed the storage industry.
The Flash Revolution
The proliferation of NAND flash memory, colloquially referred to as solid state disk or SSD, has changed the storage
industry profoundly. Where it has been previously necessary to short-stroke many 15k RPM fiber-channel disks to
achieve a high number of I/O operations per second, the same performance characteristics, plus potentially
lower- access time and less congestion on disk, make flash memory a very attractive solution. Very high I/O
performance can now be achieved in smaller, less power-hungry, and easier-to-cool solutions either inside the
database host or externally connected.
Interestingly, consumers benefit from flash storage in a similar way to enterprise customers, although the way
the storage is actually presented to the hardware differs greatly between the two customer groups. Most enterprise
offerings for flash-based memory, which in reality is NAND flash, fall into the following categories:
Connected internally to the server or blade
As 2.5-inch or 3.5-inch solid state disk
As a PCI Express card
Externally attached
Fiber Channel
Via PCIe card
Infiniband
 
Search WWH ::




Custom Search