Database Reference
In-Depth Information
■
Let's not forget that this is an academic comparison. You currently will not be able to build a system with
enough draM and backup batteries to compete for capacity in a cost-effective way with a multiple-shelf, enterprise
storage array.
Note
What is striking, though, by looking at the numbers is what is called the access time gap: less than 100
nanoseconds to access DRAM compared to 8,000,000 nanoseconds for magnetic storage is quite a difference.
You can also see from the above table that the latency for flash memory access is a lot lower than for hard disks,
but before you start to base your whole storage estate on SSD, you need to know a little more. Let's approach the
complex topic of SSDs using the categorization in Figure
3-1
as our basis.
Figure 3-1.
Attempt of a categorization of SSD
The type of SSD is important, but there is little to say other than that (D)RAM-based SSDs do not have a
significant market share anymore. Practically all SSDs today are NAND-flash based. This leads to the next item in the
list: DRAM is volatile memory, which means that if there is a power cut, then all the data is potentially lost. Vendors
realized that early on and added batteries to their DRAM-based solutions. NAND-based flash memory does not
exhibit the same behavior. Please note that the following sections focus on NAND-based flash memory.
You already read about the access paths to the NAND storage; they are listed again for the sake of completeness.
What you need to remember is that accessing SSD via PCIe is the quickest path to the storage. This is followed by
Remote Direct Memory Access and finally by Fiber Channel.
The type of memory cell bubble denotes how many bits a cell in a NAND-based SSD can store. To better
understand the significance of that sentence you need to know that the SSD's base memory unit is a cell. In other
words, information is stored in a non-volatile way in an array of cells. It is the amount of information that can be
Search WWH ::
Custom Search