Massive
hard disk archives can reduce running costs by powering down most of the disk -
based on useage patterns.
When a hard disk is powered it takes about
10 to 20 seconds (depending on the model) to become usable. Some enterprise -
hard drives instead offer an alternative power saving mode in which provides
faster start up -closer to 6 seconds - but smaller power savings. This strategy
reduces electricity costs but provides slow random access () to infrequently
used data. This technique - called
AutoMAID
- was pioneered by Nexsan
Technologies in 2001. Many other vendors now offer similar features in
their RAID systems, but they call the technique by differing names.
This
hard disk technique is faster than that for typical
tape libraries - which
have random access times in the range 10S to 60S. But backup SSDs will offer a
new dimension of possibilities for backup archive designers.
the
Need for Fast Startup SSDs
Like HDDS - most 2.5" SSDs
designed for the high IOPS enterprise market typically consume 2x to 3x more
power when they are reading or writing data than when they are not. And it is
unusual for SSDs (except those designed for low power embedded applications) to
support power management commands which improve much on this.
If we
want to design an affordable massive array of SSDs for an archive systems
- we would like the "standby" power consumption to be more like 10x
to 50x lower than the standby mode.
Let's look at the feasibility
of this. The switch side of that is easy - and already exists. But we need to
ask the question - how long does a server class flash SSD take to become
operational?
(OK I realize that this is an extremely simple view -
because if you really wanted to do this you may be able to optimize the power
management - but let's see if it would be worthwhile in this extreme case.)
In
the current state of the SSD market - the time taken for an SSD to become
operational from the time that power up reset is applied varies considerable
depending on the type of
internal cache,
flash memory (MLC takes longer than SLC) and
controller used. Here
are some typical figures for
2.5" SATA SSDs.
I've assumed the use of SATA
SSDs - even though they have longer power up times and consume more power
than PATA SSDs
because if you are planning to build a cabinet which houses hundreds of SSDs -
it's more realistic to assume they are distributed. This is not a
notebook.
Today's SSD market does not pay a premium for fast power
up operation, But there is no technical reason why the power up time for a
skinny flash SSD should take much longer than 100mS.
Below this figure
- the requirements for fast start-up begin to conflict with other desirable
characteristics such as
tight data management which helps the device power down reliably in the
event of sudden
power loss.
the indivisible minimum (or atomic) start-up time cannot be below about
10x a single write-erase cycles - because there may be pending internal
housekeeping operations which have to be completed - before it's safe to
initiate new R/W activity.
At one extreme end of this
characteristics the user would get an enterprise bulk storage SSD array
with the following system characteristics
power consumption - upto 50x lower than for the same capacity implemented
by standard SSDs or low power HDDs.
random R/W IOPS and throughput in the powered and ready state - which is
similar to that for average notebook SSDs
power up to ready time (for any random SSD) on the order of 50 to 100
milli-seconds. When powered up the device controller would keep the device
powered for a minimum period of seconds set by the system.
high availability fault tolerant design
Summary
A
new class of high integrity, low standby current, fast power up to ready
flash SSDs (50mS to 100mS) would simplify the design of
high density petabyte
SSD arrays for bulk archiving and backup applications - because most of the
SSDs in these arrays could remain unpowered for most of the time. It's possible
to design such power management for SSDs using today's technologies. There are
some overlaps with the requirements for fast boot notebooks too - but the
greater data integrity requirements of SSD backup means these will probably be
different products.
.
.
...
.
Do I really have to get going?
.
this way to the
Petabyte SSD - This article describes the future storage architecture of
the datacenter, explains the economics of SSDs replacing HDDs for bulk storage
and suggests a roadmap for getting there. .
SSD Data Recovery
- includes articles and news related to recovering data from faulty or damaged
SSDs. .
SAS SSDs -
includes a timeline of the SAS SSD market - and lists significant vendors. .
the SSD
Reliability Papers - links and abstracts of articles related to the subject
of SSD reliability and data integrity. .
Surviving
SSD sudden power loss - this article surveys SSD power down management
across all the SSD architecture types in the market today. .
the problem
with Write IOPS in flash SSDs - this classic article helps you understand
why all SSD benchmarks incorrectly suggest you're going to get much higher
performance from some types of flash SSDs than you will actually see in your
application.. .
HA
SSDs - is a new article and directory about fault tolerant enterprise SSD
systems. .
Fast Purge SSDs
- is an article which includes a directory of vendors who design SSDs which can
self destruct or quickly and securely erase flash SSD contents (typically in a
fraction of a second) to prevent data getting into unwanted hands. .
RAM Cache
Ratios in flash SSDs - all SSDs are in one of these categories - which also
predicts operational characteristics. .