Samsung 950 pro vs Crucial BX100 in RAID-0

[biostud]

What would you choose of the following?

Samsung 950 pro 512GB is $349

4x Crucial BX100 250GB for $340 RAID-0

2x Crucial BX100 500GB for $320. RAID-0

 

[CiPHER]

Do not buy BX100 - buy a good SSD such as the Crucial MX200. The BX100 uses a totally different controller like Silicon Motion and is an unprotected SSD, the Crucial MX and M series are protected by power-safe capacitors and RAID5 bitcorrection.

If you want to RAID0-them fine. It will roughly double your speeds of sequential I/O and random I/O (except blocking random reads) but even an SSD of double the speed will not translate into double the speed experience - your CPU will be a major bottleneck. So the actual speed increase is only modest. For loading games the additional speed would be on the threshold of being noticeable so it would have at least some benefit though.

But then go for the MX200 not the budget BX100 that is an SSD no one should buy.

 

[Hellhammer]

Do not buy BX100 - buy a good SSD such as the Crucial MX200. The BX100 uses a totally different controller like Silicon Motion and is an unprotected SSD, the Crucial MX and M series are protected by power-safe capacitors and RAID5 bitcorrection.

The MX200 does not have full power-loss protection.

http://www.anandtech.com/show/8528/micron-m600-128gb-256gb-1tb-ssd-review-nda-placeholder

OP, RAID-0 will only improve performance of large transfer sizes and high queue depths because the drives are still bottlenecked by AHCI. As the 950 PRO utilizes NVMe, it provides better performance at low queue depths, which are more common in client workloads. Ultimately it's question of performance vs capacity.

 

[Berryracer]

The BX100 is one of the crappiest SSDs. Not saying get the Samsung but just not the BX100.

SanDisk Extreme PRO FTW

 

[CiPHER]

The MX200 does not have full power-loss protection.

http://www.anandtech.com/show/8528/micron-m600-128gb-256gb-1tb-ssd-review-nda-placeholder

We have talked about this before i think. Anandtech does not properly understand how power-loss protection works in my opinion. Not in the past and still you guys do not fully understand why power-loss protection is so crucial (pun intended) and Samsung can do without.

But you are correct in the sense that Crucial has only partial power-loss protection. It will still lose the contents of the DRAM buffercache. But that is not any problem; this is also the case for all harddrives used in the last 20 years. Filesystems of the first generation (FAT16/FAT32, Ext2, UFS) are vulnerable to the loss of the DRAM contents when write buffering is enabled. Without write buffering, harddrives can only write at 1MB/s. Starting from the 2nd generation of filesystems (NTFS, Ext3/Ext4, UFS2+SU) the loss of the DRAM contents is not a problem as long as FLUSH commands are adhered to. The latter is where all the fuss is about. SSDs can corrupt recent writes even across flush commands. So the write has been acknowledged by the SSD, but will still be lost some cases of unexpected power-loss.

Samsung does dirty tricks by implementing journalling on the mapping tables, so that it can ignore flush commands; it just rolls back the entire LBA storage to a consistent state in the past. This requires specific firmware, but allows omitting the use of hardware power-safe capacitors.

Crucial has decent hardware power-safe capacitors and thus its protection is much better than Samsung. It allows their SSDs to be used for more than casual storage, since Samsung SSDs theoretically can have problems when used in a RAID array -- just imagine what would happen if you have a RAID0 array and one or more SSDs revert to an earlier state, but the other SSDs in the RAID0 array do not, or revert to a different state. Then you still have filesystem corruption even though the SSD itself is in a consistent state.

OP, RAID-0 will only improve performance of large transfer sizes and high queue depths because the drives are still bottlenecked by AHCI.

What does AHCI have to do with it?

You are right though that you need higher queue depths, but that only counts for reading. Writing does not rely that much on queue depth. With just a queue depth of 2 or 4 you can queue up to 40 writes thanks to buffering. But for random reads the same trick is not possible. 4 queued I/Os means 4 concurrent reads. Since SSDs themselves already implement 1:16-way interleaving, you need about 32 queued I/Os to almost saturate a RAID0 of 2 SSDs.

But it still means an increase in random IOps, RAID0 does not only increase sequential speeds as many people think. Also thanks to very bad reviews/benchmarks on both Anandtech and Tomshardware (Achim Roos & Patrick Schmid in particular).

NVMe just decreases the latency, which is very useful, but does not change any of the above as far as i know.

 

[Redstorm]

4 x 950's RAID 0 of course.

 

[CiPHER]

I would opt 4x Crucial MX200 using only half of storage so you can utilise them as SLC SSDs instead of MLC. That would give you the best of all latency, even better than the latency of MLC SSDs like the Samsung 950 Pro (which uses MLC instead of TLC like the 840 (EVO) and 850 (EVO) do).

 

[Redstorm]

Nice, took the bait. too easy.

 

[Hulk]

Do not buy BX100 - buy a good SSD such as the Crucial MX200. The BX100 uses a totally different controller like Silicon Motion and is an unprotected SSD, the Crucial MX and M series are protected by power-safe capacitors and RAID5 bitcorrection.

If you want to RAID0-them fine. It will roughly double your speeds of sequential I/O and random I/O (except blocking random reads) but even an SSD of double the speed will not translate into double the speed experience - your CPU will be a major bottleneck. So the actual speed increase is only modest. For loading games the additional speed would be on the threshold of being noticeable so it would have at least some benefit though.

But then go for the MX200 not the budget BX100 that is an SSD no one should buy.

I've had a BX100 500GB in my laptop for about 6 months and it has performed admirably. Is the lack of power loss still a big problem even in a laptop?

 

[Soulkeeper]

If choosing between a BX100 and the MX200 i'd pick the MX.
The main selling point of the BX is it's low power usage and slightly lower price, outside of that the MX is just better. You can usually get a larger MX drive for the same price as a smaller 850pro or even evo. Ultimately these things come down to personal preference.
950pro I dunno 512GB for $349 just seems steep to me.

 

[CiPHER]

BX100 and other budget SSDs often perform very decent - that is why they are sold; they are cheaper but still have good specs.

But in my opinion - and this is truly an opinion - they should not be sold. No SSD which is unprotected should ever be sold. This is the year 2015 - and we should expect computing technology to be reliable to the degree where a computer problem where the technology is to blame, is extremely uncommon to the point of being non-existent. SSDs have all the potential to be vastly more reliable than mechnical technology like HDDs, which are unreliable because of their mechanical nature. But solid state technology such as electronics, can be extremely reliable and durable for the economical lifespan of the product. An SSD could be used for 20 years in this era, and there is no reason the product cannot be designed to perform reliably during this time.

But, because making money is more important than providing consumers with high-grade technology, crappy products are being solds by the millions. And all across the world, we consumers get to use inferior products such as OCZ and many others. The result is not that every one using these products have apparent problems. Some genuinely have no problems at all and can be quite content with their product. I consider this to be the same as smoking all your life and not getting cancer. Good thing you cheated death! The dice were in your favour - this time. But for others, the dice are less favourable, and can actually cause quite a bit of trouble.

The problem with unprotected SSDs like the BX100 is, that problems may occur without the consumer ever knowing about it. The SSD continues to function, but still there are strange errors of Windows, or a specific application or game, or boot problems, or hanging applications, or blue screens. All this while, the SSD may continue to function as if it were not to blame. But it is, because these problems can be the result of the SSD causing corruption. This can happen because these SSDs are unprotected and have neither RAID5 bitcorrection to protect against corrupted/unreadable NAND pages, or against corruption of the mapping tables (FTL) in case of unexpected power-loss. The latter happens far more than one might assume. Do not think: i have a power failure once a year so unexpected power-loss also means once a year. You can have zero power outages and still have hundreds of unexpected power-loss. Simply resetting your computer during the boot phase, a blue screen, or a loose power-cable can result in the SSD having unexpected power-loss. Even tidy shutdowns can cause this because various bugs can cause the power to be cut before the STANDBY IMMEDIATE command being processed by the SSD - which is the command to signal a tidy shutdown.

In short, i believe consumers should pay more attention to reliability than performance. The truth is, all SSDs are fast by nature because of NAND flash-memory. The differences between SSDs are quite marginal, especially for consumers since they rely more than anything on the LOWEST number the SSD puts out in benchmarks: 20MB/s of blocking random 4K read performance is the true reason SSDs feel so snappy and respond instantly to user input. Yeah.. that is right! All those high numbers are little more than bragging rights and marketing cannonfodder.

To nuance my story, the BX100 is certainly not a very crappy SSD. It work work for you and others. But it continues to have a 'window of opportunity' in which the SSD can corrupt itself. In other words, it has been designed to cause corruption and thus cause problems you do not want. The chance is just not very big. It will work just fine most of the time. But to me, that is enough, it has to work ALWAYS because consumers earn the right of getting to use reliable technology. And there is no reason this is not possible with the current state of technology. Crucial MX200 is just slightly more expensive and is a protected SSD, while others are not. Intel 320 continues to be the most reliable consumer-grade SSD out there, but it is end-of-life and expensive and not all that fast with today's standards.

So really, reliability and price should be the points to look at. Performance is not all that interesting. All SSDs perform around the same in benchmarks that are realistic to typical consumer workloads.

Amen. :hmm:

 

[VirtualLarry]

The problem with unprotected SSDs like the BX100 is, that problems may occur without the consumer ever knowing about it. The SSD continues to function, but still there are strange errors of Windows, or a specific application or game, or boot problems, or hanging applications, or blue screens. All this while, the SSD may continue to function as if it were not to blame. But it is, because these problems can be the result of the SSD causing corruption. This can happen because these SSDs are unprotected and have neither RAID5 bitcorrection to protect against corrupted/unreadable NAND pages, or against corruption of the mapping tables (FTL) in case of unexpected power-loss. The latter happens far more than one might assume. Do not think: i have a power failure once a year so unexpected power-loss also means once a year. You can have zero power outages and still have hundreds of unexpected power-loss. Simply resetting your computer during the boot phase, a blue screen, or a loose power-cable can result in the SSD having unexpected power-loss. Even tidy shutdowns can cause this because various bugs can cause the power to be cut before the STANDBY IMMEDIATE command being processed by the SSD - which is the command to signal a tidy shutdown.

Just because an SSD has power-protection capacitors, does NOT make it immune from glitches. I had some sudden power-off issues because of an incompatibility between my Delta-made Antec EarthWatts 650W PSU, and my CyberPower 550VA UPS, and when it went to switch to battery, my PC would suddenly shut off.

Anyways, my primary SSD is a Crucial M500, which as we all should know, was advertised as having power-protection capacitors.

Yet, even so, I've noticed occasional corruption. Specifically, I downloaded a file, my browser download history said the entire file was downloaded (and listed a size), but the file on-disk in the Windows 7 64-bit filesystem (native NTFS) was a "runt".

So, I figure the SSD has some sort of internal corruption in the mapping tables.

So, don't be afraid of getting a BX100, because the power-backup capacitors in the more expensive SSDs don't do all that much anyways for you.

 

[MongGrel]

Do not buy BX100 - buy a good SSD such as the Crucial MX200. The BX100 uses a totally different controller like Silicon Motion and is an unprotected SSD, the Crucial MX and M series are protected by power-safe capacitors and RAID5 bitcorrection.

If you want to RAID0-them fine. It will roughly double your speeds of sequential I/O and random I/O (except blocking random reads) but even an SSD of double the speed will not translate into double the speed experience - your CPU will be a major bottleneck. So the actual speed increase is only modest. For loading games the additional speed would be on the threshold of being noticeable so it would have at least some benefit though.

But then go for the MX200 not the budget BX100 that is an SSD no one should buy.

I'd agree.

I bought a BX-100 just for a little drive for the OS on a HTPC that doesn't get pushed hard, it works great.

It does have a X5650 in it, way overkill, but put the X5680 in the main.

But bought a larger MX-200 for to upgrade friends computer when they needed one on their main and they love it.

Still using a couple older Sammy 850 Evos in RAID0 on my main here for the OS, haven't had any problems.

They transfer in around 750+

 

[CiPHER]

Just because an SSD has power-protection capacitors, does NOT make it immune from glitches.
[..]
I've noticed occasional corruption. Specifically, I downloaded a file, my browser download history said the entire file was downloaded (and listed a size), but the file on-disk in the Windows 7 64-bit filesystem (native NTFS) was a "runt".

I am not really sure what you mean with 'runt'. But the fact that your application says the download is complete, does not mean it has been flushed to disk. Data on the SSD or HDD is only safe as being permanently stored when the subsequent FLUSH CACHE command has been completed. This normally only occurs upon filesystem metadata writes. There can be quite some time between these flush-commands, depending on I/O traffic.

Normally flush commands are very sparse because of the performance impact these have. If all writes would be synchronous - meaning each write would be followed by a FLUSH command - then harddrives would write at about 1MB/s at the most or little over 50K/s for random writes.

So, don't be afraid of getting a BX100, because the power-backup capacitors in the more expensive SSDs don't do all that much anyways for you.

I think this is a wrong conclusion, but i respect your opinion.

 

[VirtualLarry]

I am not really sure what you mean with 'runt'. But the fact that your application says the download is complete, does not mean it has been flushed to disk. Data on the SSD or HDD is only safe as being permanently stored when the subsequent FLUSH CACHE command has been completed. This normally only occurs upon filesystem metadata writes. There can be quite some time between these flush-commands, depending on I/O traffic.

Normally flush commands are very sparse because of the performance impact these have. If all writes would be synchronous - meaning each write would be followed by a FLUSH command - then harddrives would write at about 1MB/s at the most or little over 50K/s for random writes.


I think this is a wrong conclusion, but i respect your opinion.

A "runt" is a shorter-than-nominal piece of data. The term is used in ethernet networks a lot.

Anyways, I wasn't trying to say that the SSD experienced a power loss between downloading and attempting to use the file. It didn't. Just that it seemed that it was internally corrupted, possibly from the forced power-offs, causing FS corruption in Windows.

 

[Elixer]

But you are correct in the sense that Crucial has only partial power-loss protection. It will still lose the contents of the DRAM buffercache. But that is not any problem

It still results in data corruption, so, I beg to differ, it still is a problem.

;this is also the case for all harddrives used in the last 20 years. Filesystems of the first generation (FAT16/FAT32, Ext2, UFS) are vulnerable to the loss of the DRAM contents when write buffering is enabled. Without write buffering, harddrives can only write at 1MB/s. Starting from the 2nd generation of filesystems (NTFS, Ext3/Ext4, UFS2+SU) the loss of the DRAM contents is not a problem as long as FLUSH commands are adhered to. The latter is where all the fuss is about. SSDs can corrupt recent writes even across flush commands. So the write has been acknowledged by the SSD, but will still be lost some cases of unexpected power-loss.

Thing is, this is a corner case, and while I do agree that SSDs should have power-loss protection, and it should be pointed out in reviews, but, realistically, power-loss isn't the #1 factor for most people.
Lots of people have UPS systems these days, so this specific situation is mitigated. Laptops have batteries, so, those are pretty much protected from this specific situation as well.

For the OP, why are you looking at RAID 0 anyway, what is your workload?

 

[CiPHER]

It still results in data corruption, so, I beg to differ, it still is a problem.

It is only a problem for first generation filesystems, which nobody uses these days. All 2nd generation filesystems can cope with lost recent writes up to the last FLUSH CACHE command. If it loses writes beyond this boundary, then the filesystem is in trouble - especially with reordered writes. That is why you need BBU or Battery Backup Units for unsafe write-back buffercache as used on Hardware RAID adapters like Areca. You do not need BBU for a normal harddrive.

So what you say is incorrect to my knowledge.

Lots of people have UPS systems these days

No they don't. Less than 0,1% of consumers have a UPS. Even with a UPS, that does not mean you will not have unexpected power-loss. Look in the SMART data for 'Unexpected Power-Loss Count' and explain to me if you had as many power failures as the number there suggests. Then, how do you explain the high number?

Right... power outages have nothing to do with unexpected power-loss. The former is about loss of AC power, the second is about unsafe shutdowns without informing the SSD that a power loss is about to occur. Using the reset button at BIOS phase is enough to trigger this.

 

[Elixer]

It is only a problem for first generation filesystems, which nobody uses these days. All 2nd generation filesystems can cope with lost recent writes up to the last FLUSH CACHE command. If it loses writes beyond this boundary, then the filesystem is in trouble - especially with reordered writes. That is why you need BBU or Battery Backup Units for unsafe write-back buffercache as used on Hardware RAID adapters like Areca. You do not need BBU for a normal harddrive.

So what you say is incorrect to my knowledge.

You are assuming one write transaction is sufficient when you have a power loss event.
I am talking about if you have lots of data (say an archive), and if you are not on battery, then the last write transaction before the power loss event will still result in the archive being corrupted if it didn't finish writing all of it.

No they don't. Less than 0,1% of consumers have a UPS. Even with a UPS, that does not mean you will not have unexpected power-loss. Look in the SMART data for 'Unexpected Power-Loss Count' and explain to me if you had as many power failures as the number there suggests. Then, how do you explain the high number?

Either having a UPS or laptop (which have batteries) is far over your .1%, I am thinking over 60% if not more.
All my SSDs show 0 for that stat. As to what triggers an "unexpected power-loss"event, it can vary from each SSD maker.
Normally, they have a voltage monitor that can detect power loss, and when that kicks in, it dumps the last write transaction to the NAND.
Then, when normal operation continues, it checks that block to verify it (well, that is a vastly simplified version of what is going on).
We can have power-loss events in multiple stages as well, and depending on where in the pipeline the write transaction is, it can be handled differently.

Right... power outages have nothing to do with unexpected power-loss. The former is about loss of AC power, the second is about unsafe shutdowns without informing the SSD that a power loss is about to occur. Using the reset button at BIOS phase is enough to trigger this.

Hmm?
A power outage can indeed cause a "unexpected power-loss" event, as can hitting the reset button, as can the SSD going into panic mode, and most likely other corner cases as well that are specific to each firmware.

 

[Deders]

Does Raid 0 have any affect on 4k read and writes? Last I checked it didn't improve them at all which is where the NVMe interface will shine. Seeing as this is the biggest bottleneck to performance it is the biggest reason I'm excited about the 950 pro.

Also with Raid 0 you have to rely on 4 disks to be stable which overcomplicates things.

 

[Hellhammer]

Crucial has decent hardware power-safe capacitors and thus its protection is much better than Samsung. It allows their SSDs to be used for more than casual storage, since Samsung SSDs theoretically can have problems when used in a RAID array -- just imagine what would happen if you have a RAID0 array and one or more SSDs revert to an earlier state, but the other SSDs in the RAID0 array do not, or revert to a different state. Then you still have filesystem corruption even though the SSD itself is in a consistent state.

If you read the article I linked, you would know that the capacitors in all Crucial/Micron client drives only protect lower pages from corruption. Data in lower pages is old data that has been flushed ages ago, so it's considered data-at-rest. The capacitors provide absolutely zero protection for any data in the DRAM, including the FTL, meaning that Crucial is also resorting to journaling to recover the FTL in case of a sudden power loss.

From http://www.samsung.com/us/business/oem-solutions/pdfs/V-NAND_technology_WP.pdf

Capacitors for lower page protection are not needed with V-NAND because all bits, even with TLC, are programmed to a cell at once.

I think you are exaggerating the need for full power loss protection in client usage. If data corruption was a significant issue, then we would have people telling their stories and complaining about it in this very forum. It is true that client SSDs are vulnerable to power losses to some extent, but there are proper mechanisms in place to protect the FTL and data-at-rest. The boot time after a power loss may be a little longer as the SSD recovers the FTL, but that it pretty much the only "issue" that a client user will ever see.

You can always pay the premium and buy an enterprise drive if you don't trust the protection that client drives provide. In the end, client-grade hardware is always a compromise between cost and functionality.

NVMe just decreases the latency, which is very useful, but does not change any of the above as far as i know.

Throughput and IOPS are inverses of latency, so any decrease in latency will result in higher performance. That's why NVMe is more than just a marketing gimmick because it actually improves 4KB QD1 random performance substantially (SM951-NVMe is ~50-90% faster than 850 PRO). RAID-0 doesn't help with that because it only increases parallelism - it doesn't improve the minimum latency at all.

 

[bgstcola]

I would opt 4x Crucial MX200 using only half of storage so you can utilise them as SLC SSDs instead of MLC. That would give you the best of all latency, even better than the latency of MLC SSDs like the Samsung 950 Pro (which uses MLC instead of TLC like the 840 (EVO) and 850 (EVO) do).

Sounds interesting. Where can I read about that? Thanks

 

[Deders]

Is SLC enough to give it 300,000 iops?

 

[tweakboy]

Looks like your watching porn and have the biggest porn collection in the world with allt hose drives that you really dont need,, RAID is only good for bulk transfers,, other then that its same sh*T different smell.

 

[CiPHER]

You are assuming one write transaction is sufficient when you have a power loss event.
I am talking about if you have lots of data (say an archive), and if you are not on battery, then the last write transaction before the power loss event will still result in the archive being corrupted if it didn't finish writing all of it.

Alright, but this is application-level corruption/inconsistency, not filesystem or storage device (LBA) level inconsistency. The application is responsible for its own stuff. The filesystem only guarantees that application writes that are synchronous to be completed before the I/O request is finished. Asynchronous writes by the application will be finished immediately if the buffer allows, but unlike sync writes the async writes have no such guarantee.

So we are talking about different things here.

Either having a UPS or laptop (which have batteries) is far over your .1%, I am thinking over 60% if not more.

You cannot be serious about that? We are talking about ordinary consumers? You are saying 60% of consumers have a UPS? Seriously? Well if you count the laptop battery as UPS that would indeed increase the number significantly, from 0,001% to 20% maybe. But still, that is far below the number you suggest.

And just to repeat: having the power supply guaranteed does not mean your SSD will not have unexpected power-loss. That has to do with tidy shutdowns more than anything.

All my SSDs show 0 for that stat. As to what triggers an "unexpected power-loss"event, it can vary from each SSD maker.

Not really, as far as i know the STANDBY IMMEDIATE command is responsible for that. If you have more information i'm all ears.

If you read the article I linked, you would know that the capacitors in all Crucial/Micron client drives only protect lower pages from corruption.

Which is not what i disputed, is it?

The capacitors provide absolutely zero protection for any data in the DRAM

Also not what i disputed.

including the FTL, meaning that Crucial is also resorting to journaling to recover the FTL in case of a sudden power loss.

Now this is some good point you make. And a very crucial one. Because even if it is true what i asserted, that Crucials capacitors allow for the pending write to succeed not to cause corruption to the MLC 'lower pages' as you describe, the FTL/mapping tables that are cached in DRAM might still be newer than what is written to NAND. And as such, the loss of the DRAM contents does not only lose recent async writes (buffered writes) but also causes the loss of the newer FTL version. And as you say, the only valid defence against such an occurrence would be to provide journalling of the mapping tables. If this is true, then the protection is not what i anticipated and asserted. And you would be very right to set this straight. So i guess i owe you an apology - it seems i did not fully understand the situation properly.

I think you are exaggerating the need for full power loss protection in client usage. If data corruption was a significant issue, then we would have people telling their stories and complaining about it in this very forum.

Well, aren't there quite a bit of stories about strange things happening with SSDs? We all know the stories about FTL corruption where you need to provide power to the SSD for half an hour or so, to allow it to make its FTL consistent with the data again. I think we can all understand that this is due to an FTL inconsistency - where the FTL does not match the data stored on NAND, very likely due to unexpected power-loss.

If this is true, then I would assert that the lack of proper protection against power-loss introduced a 'window of opportunity' at which the SSD can corrupt itself.

Even worse, this kind of corruption is hard to detect and to many people - especially consumers - it would not be all that apparent that the SSD is to blame. Strange error messages, hanging applications, blue screens or boot problems - would we suspect the SSD right away? With the SSD fixing itself meaning that it continues to function even with LBA-level corruption, the consumer might not suspect the SSD is to blame since it is working properly again.

So in this era where technology can 'sometimes work, sometimes not' the failure modes are much more subtle than a simple 'works' or 'failed' - which is far easier to diagnose.

My own argument would be that whenever you have a computer problem with weird symptoms, you do not want to suspect your SSD because you use an unprotected SSD that still has a window of opportunity where it can corrupt itself. We simply want reliable technology in 2015 - not something that works 99,9% of the time. Sadly though, this appears to me to be the case for consumers which get to use something that mostly works, only to satisfy the business users who require near-100% reliability and are willing to pay for it. This means the manufacturer has an interest to prevent consumer-grade technology/products becoming too reliable, forcing business-oriented users to buy the more expensive enterprise-grade products which have decent protections. The lack of ECC RAM memory for consumer systems only confirm this behaviour. Since there is no valid reason why not all systems would have ECC protection. The added cost is very marginal, and ECC is used in plenty other areas of the computer system. So why not spend a fraction of added cost to have the RAM memory protected? The only reason can be that the big companies want their more expensive enterprise-grade products to be separated for the consumer-grade stuff. And it is this practice in particular that i oppose. I think all consumers should have access to reliable computing technology in the year 2015. There is no valid technical reason why does cannot be feasible.

It is true that client SSDs are vulnerable to power losses to some extent, but there are proper mechanisms in place to protect the FTL and data-at-rest.

Journalling or full power-loss protection, i do not know any other ways to protect this risk? Do you?

And while i have no detailed knowledge about every protection SSDs utilise, i strongly suspect that budget SSDs have no proper journalling feature to protect the FTL from corruption/inconsistency. If this is true, it would confirm my bias for protected/unprotected SSDs and I would assert consumers should buy only protected SSDs that have proper protection against FTL corruption. Same with RAID5 bitcorrection (RAISE/RAIN or whatever companies like to call it). Every SSD should have it. Even RAID6 (double parity) - why not? The M500 uses 1:16 parity, whlie the MX100/MX200 uses 1:128. Why not increase this to 2:128? It would only cost 2/128 of storage capacity or 1/64th. That is not all that much to increase the uBER to enterprise-levels. It is so cheap to do this, pure software, no added physical costs to produce, so why not? From a technical point of view this only makes sense. But from a marketing strategy point of view it is stupid to allow SOHO/business-users to use cheap consumer-grade products which have less profit margin.

Throughput and IOPS are inverses of latency, so any decrease in latency will result in higher performance. That's why NVMe is more than just a marketing gimmick because it actually improves 4KB QD1 random performance substantially [..] RAID-0 doesn't help with that because it only increases parallelism - it doesn't improve the minimum latency at all.

Alright now i understand what you mean. But AHCI does not have any direct link with RAID0. You only meant to say that while RAID0 increases all performance specs, it cannot increase blocking random reads (commonly known as 4K QD=1) - fair enough!

 

[CiPHER]

Does Raid 0 have any affect on 4k read and writes?

RAID0 doubles Random IOps just as it doubled sequential I/O. But, there is one exception. It will not increase blocking random reads, commonly referred to as 4K QD=1 or simply '4K'. This is only true for reads though - writes even with qd=1 can be improved with RAID0 thanks to write buffering. But with reads this is not possible.

This is why all SSDs score about the same for this benchmark spec - about 20MB/s. SSDs themselves use RAID0 internally on multiple levels. In fact, you can consider a single SSD to be a RAID0 of 16 NAND devices, like a 16-disk RAID0. The performance characteristics are very similar. You require multiple queue depth to saturate the SSD for random reads, just like a RAID0 array would. That is the 4K QD=32 number in AS SSD and CrystalDiskMark.

Last I checked it didn't improve them at all which is where the NVMe interface will shine.

NVMe does increase the most important performance spec; 4K blocking reads. This is the most important performance spec for consumers and one that RAID0 (both internal & external) cannot accelerate. So NVMe for consumers is very useful. Though still, a faster SSD does not translate into vastly improved speed experience. A single SATA/600 SSD with AHCI-interface is already very fast to let the CPU be the bottleneck most of the time.

Sounds interesting. Where can I read about that? Thanks

Not sure where you can. But using the MX200 250GB as SLC SSD is simple: simply do not write more than half the LBA/storage capacity. You can force this by using partitions and not partition more than 50%. You might want to partition 45% to provide some overprovisioning as well. So 110GB partition would be good. Then you have a very cheap killer-SSD that operates as SLC SSD at very marginal consumer-grade cost. I like it! :awe:

Is SLC enough to give it 300,000 iops?

Sheesh, why you need 300.000 IOps? You going to run 2000 websites on a single SSD or what? Seriously, the SLC just provides a marginal decrease to latency which improves performance no MLC/TLC SSD can. It also improves lifespan if you write to the SSD a lot, for example when the SSD is used as caching SSD like with Intel SRT or ZFS L2ARC. It does not provide all that much benefit for really ordinary consumers. Those want the SSD to be as cheap and large as possible.