free web stats

SSD Endurance Standardized – MTBF Has to Go

by: admin Friday, October 8th, 2010

JEDEC has published new standards for measuring lifetime and reliability in SSDs. The new standards, JESD218 (Solid-State Drive (SSD) Requirements and Endurance Test Method) and JESD219 (Solid-State Drive Endurance Workloads), are intended to make it easier for businesses to choose the best products for their needs.

The idea is to even out the playing field for solid state drives by using measurements that are more suitable for SSDs than the existing MTBF (Mean Time Before Failure) standard, which has previously been used as a statistical measure of life span for hard drives. These are however not equally well suited for flash drives, which are based on entirely different technology.

Since SSDs are exposed to various levels of load depending on different types of usage, the new standards are defined according to two application classes: Client and Enterprise.

JESD218 creates a measurement of endurance called TWB, which represents the number of terabytes of data written to the tested drives. This can be used to compare different flash drives within the same application class.

JESD219 create metrics on workload management, where the workload is standardized.

« | Home | »

8 Responses to “SSD Endurance Standardized – MTBF Has to Go”

vt Said:

It is actually not difficult to calculate the maximum amount of information written to SSD. The upper limit in terabytes is (Size of your SSD in gigabytes)*4.9. For more realistic life expectancy divide that value by 2. That would work for MLC drives. For SLC drives the multiplier would be 90.

Apparently if you care about the price of total data written to SSD, SLC would be a better choice (way less expensive when you measure the amount of data written, not the maximum capacity).

Consider the following example:
60GB Intel X25-M SSD (MLC) – 240$
32GB Intel X25-E SSD (SLC) – 420$

The total expected safe amount of non-cached data which can be written to 1st SSD is about 60*2.5 = 90 TB (terabytes). That would give us a price of about 2.7$ per terabyte written.
The second SSD, despite having lower total capacity can withstand more data written to it. The safe amount of non-cached data would be 32*45 =1440 TB (terabytes). That gives us a price of about of about 0.3 $ per terabyte written (ONLY!). That means that with high amounts of data written you will need to replace 16 such MLC SSD drives instead of 1 such SLC.
In this example you can clearly see that if you care about the amount of data written, the second SSD, while having higher price and lower total capacity would be actually 9 times cheaper for your heavy data processing tasks because of much longer actual lifespan under heavy load. Same rule applies to examples with comparable capacity SSD (for the example above i just selected SLC and MLC SSDs in close price range).
That is why if you need a SSD drive for tasks with heavy write loads, even at home, in the long run purchasing an enterprise class SLC SSD with even 4x price per total capacity compared with the respective MLC SSD would be actually a way cheaper and better choice (over MLC SSD).

Comment made on October 8th, 2010 at 4:48 pm
admin Said:

Thanks vt, those are VERY interesting calculations. On the other hand 99% of first time SSD buyers will no doubt end up with an MLC drive anyway. Price tags tend to speak for themselves and for a laptop of simple workstation capacity is also a major factor (especially in laptops).

For servers it’s of course a whole other matter. But wouldn’t you say that MLC drives can still be an alternative, considering the entire (relatively short) life cycle of the average PC?

Comment made on October 11th, 2010 at 3:33 pm
vt Said:

MLC SSDs are probably good enough for “normal” (probably most) users who don’t work with high amounts of frequently written data (where the task is just to store most of the information without erasing it for a long time). Since they don’t need to write a lot, the lifetime expectancy of their MLC SSDs would be on average way longer than their respective “life cycle”, i.e. they would just purchase a new PC in a few years before the lifetime of their MLC SSDs runs out – and at the same time they would get superior data access (reading) speed. Those “average use” numbers make it profitable for MLC SSD manufacturers to give 1-2 years of warranty (from the initial use of their SSDs) without getting much noise over separate cases of decreased lifetime of their products.
That makes MLC drives on average cheaper for “normal” users (and almost an ideal high-performance storage choice for laptop users) when it comes to PRICE/AMOUNT OF DATA STORED ratio within their normal life cycle.
So, the answer is for SSD MLC storage (when it comes to performance numbers):
VERY fast external removable media (as a faster alternate with a higher capacity to your USB flash drive) – best choice
Laptops – best choice
Normal workstations under low data processing load – best choice
Professional workstations, read-only (mostly) data access tasks – best choice
Professional workstations, intensive data processing – poor choice
Server read-only storage – best choice
Server intensive data processing – worst choice

Comment made on October 12th, 2010 at 3:18 am
admin Said:

Great categorization. It would be really interesting to delve a little deeper into that subject, and try to find the scale tips in favor of SLC/MLC on the cost/endurance scale. Are you by any chance available for a guest post? :)

Comment made on October 12th, 2010 at 2:56 pm
vt Said:

I’m usually busy with various things and i have only a few articles made elsewhere.
For now i just write random replies about some things which are well-known but usually not published.
As long as the facts are available in the form of comments or article, there is actually not much difference.
The facts i outlined here are nothing out of ordinary, they are well-known for tech workers of the respective SSD manufacturers and server experts, but such information is usually suppressed by PR departments as it could potentially lower their sales. But in their stupid efforts to get more money they just distort the real facts about the “new” technology.

Comment made on October 14th, 2010 at 4:38 am
calc Said:

60*2.5!=90

Comment made on January 28th, 2011 at 1:45 am
vt Said:

Indeed, my mistake, in that calculation that would make 180, with the price of 1.33$ per terabyte written. I didn’t replace 1.5 multiplier (for older MLC SSDs) with 2.5 for the newer ones.
Thank you for pointing this out.

Comment made on January 31st, 2011 at 2:20 am
vt Said:

and the multiplier is 3 for newer SSDs based on SandForce controller for random writing patterns.

Comment made on January 31st, 2011 at 2:29 am
 

Leave a Comment