For SSD Experts, request your input

[lightp2]

As I read another thread, I can't help but wonder about some things I posted a while back so I am going to repost again.

My position : I am neutral and in this post requesting information and confirmation. Not really arguing about policy.

1. In the other thread, the following argument was presented By one camp

P1 - Warranty for X years. Then some claim that warranty is restricted to WITHIN SPECIFICATION. Meaning if you wrote over the limit for the SSD, you are also not qualified for RMA.

here lies my problem. --> 1st I apologize for short on effort. For consumer SSD models, the only place I can find some guesstimate is the result of Intel actually putting out Datasheet for others to see.

I could not find datasheet in such detail for other brands/models. So for some who argue with such point, is it possible we can get access to such detail (meaning on the level of detail similar to the Intel datasheet) so that we can get idea exactly how much endurance is expected.

For example, I frequently see only 3-years warranty stated for SSD products. But never how much it is expected to withstand. So if I purchased from authorized SSD retailer and subjected it to 24x7 heavy database writes resulting in eary failure, will SSD companies honor the warranty RMA? or will they point me to the SLC enterprise drives and say "next time buy this model" ?

2. Pertaining to the frequently paraded SSD benchmarks, I have posted question before so I am going to post again.

2a. According to publicly posted X25-V datasheet, at QD=32, 4K-Random write is UP-TO 2.5K IOPS.

2b. Yet we can see reviews/benchmarks posting value much greater than that.

Deduction -
1. Datasheet not correct ?
2. Wrong sample for review ?
3. Overall system cache + drive caches raise the value ?
4. Benchmark program how you derive value?
5. Other hidden combination of factors affecting the value?

Note again, I am not criticizing, am just looking for concrete info.

 

[aaronspink]

here lies my problem. --> 1st I apologize for short on effort. For consumer SSD models, the only place I can find some guesstimate is the result of Intel actually putting out Datasheet for others to see.

I could not find datasheet in such detail for other brands/models. So for some who argue with such point, is it possible we can get access to such detail (meaning on the level of detail similar to the Intel datasheet) so that we can get idea exactly how much endurance is expected.

The only 3 drives I know of with specific endurance under specific conditions publish are the 2 Intel drives and the Intel/Hitachi collaboration drive. The only other drive I've seen a write endurance spec for is the Seagate pulsar which is someone convoluted at 10.5K 4K IOPS at 70% duty cycle with 65/35% R/W for 5 years which works out to 1.66 Petabytes. But its pretty convoluted.

Most vendors you are lucky if you get an actual IOPS rating, you are even luckier if they give you multiple IOPS ratings, you are even luckier if they give you the parameters/conditions for those IOPS ratings. Even someone like STEC doesn't have actual detail on IOPS or endurance.

For example, I frequently see only 3-years warranty stated for SSD products. But never how much it is expected to withstand. So if I purchased from authorized SSD retailer and subjected it to 24x7 heavy database writes resulting in eary failure, will SSD companies honor the warranty RMA? or will they point me to the SLC enterprise drives and say "next time buy this model" ?

Depends on the company and their policies.

2. Pertaining to the frequently paraded SSD benchmarks, I have posted question before so I am going to post again.

2a. According to publicly posted X25-V datasheet, at QD=32, 4K-Random is UP-TO 2.5K IOPS.

2b. Yet we can see reviews/benchmarks posting value much greater than that.

A lot of it depends on what they are actually testing. For example, how large of a span are they testing? 50MB? 100MB? 8GB? full drive? The effect that this can have on performance for any storage device can be pretty large. As an example the IOPS difference for an X25-M between an 8GB span and a full span (80/160GB) is 8.6k/6.6k vs 300.

These performance differences have to do with the mapping limitations within the device, the coalescing limitations and the caching limitations.

For the mapping limitations, within an 80GB flash drive there are 167,187,500 different 512B locations. A pointer to a location requires 28b or 25b if you limit yourself to 4K blocks. Which works out to ~100MB to map all the locations. As the storage increases so does the mapping table. You also have to ensure that the mapping table can be written to flash on any power failure. That either requires battery backup, super caps, or limiting the amount of data you have to write.

Supercaps have their own issues to deal with as far as reliability and cost. AKA, what do you do if your supercap limits the drive reliability well below every other component (1000 hours @ 70c at $15 per in one example). Batteries, we won't go there. If you only have to stay alive for a limited amount of time you can get away with fairly basic industrial grade solid caps which will effectively outlast everything else but have fairly limited energy capacity.

So there are a lot of tradeoffs made between costs, reliability, and performance which can have different effects on performance with different working set sizes.