TRIM support?

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
49,350
You will all stop the personal attacks or be banned.
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
Yes, the sequential IO speeds will usually be faster in RAID than with a single bigger SSD.

But the thing is, most people do not need sequential speeds in excess of 400MB/s read, and in fact will not notice the difference in everyday usage.

What most people WILL notice is the hassle of dealing with a RAID-0 volume as compared to a single large SSD. In addition, the RAID-0 volume will be a lot more likely to fail and lose data than the single large SSD.

So, as I wrote before, unless the OP has a specific need for the performance of the RAID-0 volume of SSDs, he will probably have a better experience with a single large SSD.
it's hardly a hassle to deal with a raid0 volume. it's a good learning experience. plus it costs less. with ich10r, it's as simple as setting the controller to RAID, hitting it on startup, creating the volume, and that's it. windows will treat it perfectly okay.

the area where it will help a lot is in backups. everyone on a single drive should be backing up their volume, and with a raid0, the backup becomes much faster since it will actually utilize the seqential or a higher qd that the raid0 will be capable of.

if this were a casual tech site then yes, I would recommend the higher capacity drive, but on [H] I don't think anyone here would have too much trouble working with the raid0, and it is a valuable experience in learning how storage and raid levels work.
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
the area where it will help a lot is in backups. everyone on a single drive should be backing up their volume, and with a raid0, the backup becomes much faster since it will actually utilize the seqential or a higher qd that the raid0 will be capable of.
That is not much of an advantage, since all decent backup routines are automated, and usually done at off-hours anyway. And since the backups are usually only done on differences, the RAID speed does not help much anyway.

Compared to the hassle of RAID, and the added chance of failure of using RAID, it is not worth it for most people. Only if someone has specific performance needs that require a RAID of SSDs, or as you say, if someone wants to learn about RAID, then it may be worthwhile. But for most people, even power users, it is better to just have a single large SSD.
 

Computurd

Gawd
Joined
Dec 19, 2008
Messages
956
You should not say always. There are some SSDs that perform significantly worse when no pages are marked invalid. For example, look at tweaktown's reviews, where you can see that the 75% full bar (light purple) is less than half of the "empty but dirty" bar (green) for a number of the SSDs that they have tested. They do not have a 100% full bar, but that would be significantly lower than the 75% full bar for many SSDs. And without TRIM, the drive will eventually be operating at 100% full all the time, even if you are currently only using a small amount of the RAID 0 space.

http://www.tweaktown.com/reviews/4179/corsair_force_3_120gb_solid_state_drive_review/index10.html
i can see your point, but this is a gross exaggeration. The differences in scores between a drive that is 25 percent full and 75 percent full is around ten percent. As a matter of fact if you go through and do the math, the average of the difference between 25 percent full and 75 percent full on the drives in that review is exactly 10 percent.

So, a drive that is 75 percent full will perform, on average, 10 percent slower than a drive that is 25 percent full.

You would be hard pressed to find many who would use SSDs at under 25 percent of their capacity. as a matter of fact, most use them around 50 percent full, so the difference in performance gets even smaller.

how does this apply to your comment that raid is as slow as full devices? even if so, there isnt much difference than an ssd that is full 25 or 50 percent to one that is full totally. less than 10 percent per drive even without raid. after that the differences get smaller. as a matter of fact, four of those ssds @ 75 percent full are FASTER than at 50 percent full. Also, the difference between 50 percent full and 75 percent, if you dont even take into the four that are faster, is an average of 5.5 percent.

At most you would be looking at a "full drive" being 15 percent slower than a drive that is full 25 percent.
IF you were comparing a full drive to a drive that is 50 percent full, the difference would probably be roughly five percent.

so you have 2 drives (or many more) that are running 10 percent slower EACH because they do not have TRIM.
but simple math tells you that you are coming ahead easily. if you have two drives performing at 90 percent of the speed of one device (which normal user would be 25-50 percent full) , then you have with two drives the performance of 1.8 SSDs. if you have 3 drives it is comparable to 2.7 and so on so forth.
bear in mind that is worst case scenario.
it is called scaling, and it is the most basic of SSD raid principles. even if you lose 10-15 percent per device, you are coming out massively ahead.

the argument that is made is that a normal users drive is empty. that is simply not true :)

Once all the LBAs have been written at least once, the drive is basically full, even if the filesystem on the SSD contains no files.
even if so (and this performance degradation is exaggerated a bit too) then the drives still will be faster in raid than single device (as evidenced in simple math above^^)
if one were really that worried about a few percent, minimal OP would easily make this a total non-issue.
 
Last edited:

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
I think you missed my point, since that is all irrelevant. I only claimed that "always" was incorrect. All it takes is a single counter-example.

Additionally, your computation is incorrect, since you used 75% when you really need to use 100% (i.e., disable TRIM, fill up the drive 100%, delete the files, then run the test) which unfortunately, tweaktown does not measure.
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
I think you missed my point, since that is all irrelevant. I only claimed that "always" was incorrect. All it takes is a single counter-example.

Additionally, your computation is incorrect, since you used 75% when you really need to use 100% (i.e., disable TRIM, fill up the drive 100%, delete the files, then run the test) which unfortunately, tweaktown does not measure.
the gap from 25 to 75 wasn't that big. I can't imagine the gap from 75 to 100 is that much more substantial.
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
the gap from 25 to 75 wasn't that big. I can't imagine the gap from 75 to 100 is that much more substantial.
Well, it seems you need a better imagination. There is a reason the enterprise SSDs reserve a lot more space than 7% like the consumer SSDs. That last 10 or 20% can make a difference.

Besides, AGAIN, it is irrelevant to my point, which is that ALWAYS is incorrect.
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
Well, it seems you need a better imagination. There is a reason the enterprise SSDs reserve a lot more space than 7% like the consumer SSDs. That last 10 or 20% can make a difference.

Besides, AGAIN, it is irrelevant to my point, which is that ALWAYS is incorrect.
so far that article does nothing to support a scenario where the solo drive with the same data beats out the raid array. at no point did their score ever go below 50% of the fully clean drive even. if you have actual data for a raid array that was filled 100%, emptied, and garbage collected, and somehow still performed less than a solo drive, post away.
 
Last edited:

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
so far that article does nothing to support a scenario where the solo drive with the same data beats out the raid array. at no point did their score ever go below 50% of the fully clean drive even. if you have actual data for a raid array that was filled 100%, emptied, and garbage collected, and somehow still performed less than a solo drive, post away.
Please try to follow the simple logic:

1) On several of the SSDs, the 75% full benchmark was less than half the speed of the dirty/empty benchmark

2) Without TRIM, a used SSD will eventually behave as 100% full (which is slower than 75% full) even if the filesystem is empty

3) Therefore, when the filesystem is empty, some of the SSDs will have used performance (100% full, no TRIM) less than 50% of the empty / dirty with TRIM performance

4) A RAID-0 of two used SSDs with less than 50% performance of an empty / dirty TRIMmed SSD will bench slower than a single empty SSD with TRIM (and likely even slower than the equivalent double-capacity SSD)

5) A RAID-0 is NOT always faster than a single SSD
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
Please try to follow the simple logic:

1) On several of the SSDs, the 75% full benchmark was less than half the speed of the dirty/empty benchmark
well if you do the math it's actually only one of the drives, the agility 3 (which has the interesting glitch of somehow being faster when 75% full than 50% full)

for the dirty/empty, they've given no indication they allowed the drive to trim. they say that either trim or garbage collection can be used to restore the performance, but they don't say which they did. considering they are trying to show you the stats for dirty drives, it's very possible they didn't trim the drive after deleting the 75% fill of data.

on top of that, you're using the 0% numbers as your test line that the dirty drive is easily beating in every scenario except the agility 3, which it basically ties there. but who are these people with empty solo drives trimming? did they just not install anything to the drives at all? put any amount of data on those drives around 25% and you've already lowed the gap so much that even the 75% full vs the 25% full will never dip below 50%. with over-provisioning, the 100% fill is going to be very close to 75% leaving there no real gap at all. the worst case for agility 3 for a 25% drive to a 75% drive is 41k to 33k, nowhere near the point where a solo drive would be faster.

assuming you're going to use the same amount of data on either setup near 25% capacity, you will have better performance on the raid. even the agility 3 pulls it off. unless of course your goal is to keep the drives completely empty, in which case an agility 3 drive will match the performance of your single drive in your vantage benchmarks if you somehow cause the drive to be 100% full even though it has garbage collection and that wouldn't happen.

2) Without TRIM, a used SSD will eventually behave as 100% full (which is slower than 75% full) even if the filesystem is empty
you have no numbers on what the difference between 75 and 100 is for these drives in this setup so we don't actually know what the difference will be, but it can't be much

3) Therefore, when the filesystem is empty, some of the SSDs will have used performance (100% full, no TRIM) less than 50% of the empty / dirty with TRIM performance
only one drive on this chart matched 50%. none of them significantly go under. none of them were actually tested in a raid config either where the controller would pick up some of the slack too, pushing it much over in terms of performance.

4) A RAID-0 of two used SSDs with less than 50% performance of an empty / dirty TRIMmed SSD will bench slower than a single empty SSD with TRIM (and likely even slower than the equivalent double-capacity SSD)
it will match it. and as soon as you write even 25% fill to your single drive, you just lowered the performance of that single drive so much that the raid array will now be much faster even at 75% to 100% fill.

5) A RAID-0 is NOT always faster than a single SSD
hey I guess you're right. in the worst case scenario (you have an empty drive that you're not actually using for anything), the ocz agility 3 (async nand on a sandforce controller) will only *match* the performance of a dirty empty drive. and as soon as you write a mere 60gb to your 240gb drive (hey you just paid about $500 for this thing, you are going to actually put data on it, right?), the agility 3 will beat it.

conclusion: if you intend to purchase ssds and not even use 25% of the drives capacity, then don't bother with a raid array. you will only match the performance of your single drive in trim that you aren't actually using. unless of course, you don't actually write more than 25% of the drive at any given point, so the garbage collection prevents the drive from becoming 100% dirty in the raid array. this would make the raid array faster as well. and as soon as you go to fill 25% of the trimmed drive, you will exceed performance in the raid. you won't ever fall under from the data we have available, you can only match and exceed what a single drive is capable of.

so I guess if someone asks me if they should get a raid array or a single drive, I'll just ask them if they're going to fill 25% of their ssd. when they respond yes, I'll tell them to go for the raid if they have the controller to handle it.
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
for the dirty/empty, they've given no indication they allowed the drive to trim.
I'm not sure what you mean by "allowed the drive to TRIM".

The TRIM commands were sent to the SSD. They describe the procedure in depth in the article they link to about their methodology for that test. dirty / empty is TRIMmed (they delete the data and empty the trash under Win7)
 

Computurd

Gawd
Joined
Dec 19, 2008
Messages
956
) On several of the SSDs, the 75% full benchmark was less than half the speed of the dirty/empty benchmark
here is the deal, you are trying to compare the performance from a empty but dirty drive, to a drive being full. that is irrelevant in the argument that raid is slower than a single device. You are assuming that a user would even use a SSD totally blank.
heres the deal: no one uses their SSD in a empty but dirty state. They use it in a partially filled state.

The differences between 25 percent (the lowest realistic normal drive capacity for a normal user) and 75 percent is small, very small.

The difference between 50 percent and 75 percent is even smaller. the amount of performance degradation is getting smaller each level of fill that you encounter. It is 5% less of a performance drop from 50 to 75 percent full.

so logic would dictate that a 100 percent full device would have either the same, or lower loss of performance ( 3-5 percent) compared to a 75 percent full drive.

the point is, there will be a very small difference between 75 percent full, and 100 percent full. the differences get smaller with each fill level.

2) Without TRIM, a used SSD will eventually behave as 100% full (which is slower than 75% full) even if the filesystem is empty
not always. even if that were a 100 percent true statement (which it is not), the difference between the normal cacpacity a user leverages and the 100 percent full is small. so do not compare empty to full. no one uses empty ssds. that negates the point.

3) Therefore, when the filesystem is empty, some of the SSDs will have used performance (100% full, no TRIM) less than 50% of the empty / dirty with TRIM performance
no one uses empty ssds. the difference between the 25/75 percent full compared to the empty/dirty is nearly as large as the margin between clean empty and full. still comparing empty to full. try comparing empty to 25/50 percent full.

i will not reply to the further examples given. they all come back to comparing empty drives.
 
Last edited:

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
here is the deal, you are trying to compare the performance from a empty but dirty drive, to a drive being full. that is irrelevant in the argument that raid is slower than a single device. You are assuming that a user would even use a SSD totally blank.
No, that is not what I am assuming. What I am assuming is that SSDs work the way they work with and without TRIM, and then apply basic logic. It is not difficult to understand.
 

Computurd

Gawd
Joined
Dec 19, 2008
Messages
956
No, that is not what I am assuming. What I am assuming is that SSDs work the way they work with and without TRIM.
here is the main issue that I see here:

too much assumption. You are stating something as fact, that has not been proven. you have no reliable data to back up your claims. If you do, please link.

you have no data on 100 percent full drive performance in a comparative analysis to a raid array, or any other drive for that matter.

I would think it would not be good advice to tell members to not do something, with little to no data, much less argue a point. i would not advise someone to either do, or not do, something unless i had some solid data to back my claims.

even the assumptive data points to this advice/assumptions that you are making as being incorrect.
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
here is the main issue that I see here:

too much assumption. You are stating something as fact, that has not been proven. you have no reliable data to back up your claims.
No, I am making minimal assumptions.

Try to follow the simple logic. A claim was made that RAID-0 of two SSDs was always faster than a single larger SSD with TRIM.

I linked to data, and used simple reasoning, to show a counter example where RAID-0 of two SSDs would not be faster than a single SSD with TRIM.

Therefore, RAID-0 of two SSDs is NOT always faster than a single SSD with TRIM.

Please try to think about this and understand the basic logic. No need to post more nonsense arguments with bogus assumptions.
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
I'm not sure what you mean by "allowed the drive to TRIM".

The TRIM commands were sent to the SSD. They describe the procedure in depth in the article they link to about their methodology for that test. dirty / empty is TRIMmed (they delete the data and empty the trash under Win7)
I read that link before I posted. I ctrl+f'd for trim on all 7 pages... it wasn't there. they really don't give indication that the drive was trimmed at all but I guess we are supposed to assume that since they are on windows 7
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
I read that link before I posted. I ctrl+f'd for trim on all 7 pages... it wasn't there. they really don't give indication that the drive was trimmed at all but I guess we are supposed to assume that since they are on windows 7
They specifically said they deleted the files and emptied the trash. With Windows 7, when you empty the trash, the TRIM commands are sent.

Besides, the data makes no sense if there was no TRIM. After filling it up to 75%, how could the benchmark get much faster after deleting the files, unless there was TRIM or secure erase (and they would certainly not call a SE'd SSD "empty / dirty")? Just deleting files without TRIM or SE, the SSD does not know that the sectors have been freed -- the SSD would still have the pages corresponding to at least 75% of the LBAs marked as valid, despite the files being deleted. Unless they were TRIMmed, which they obviously were.
 

Computurd

Gawd
Joined
Dec 19, 2008
Messages
956
No, I am making minimal assumptions......

Try to follow the simple logic. A claim was made that RAID-0 of two SSDs was always faster than a single larger SSD with TRIM.

I linked to data, and used simple reasoning, to show a counter example where RAID-0 of two SSDs would not be faster than a single SSD with TRIM.

Therefore, RAID-0 of two SSDs is NOT always faster than a single SSD with TRIM.

Please try to think about this and understand the basic logic. No need to post more nonsense arguments with bogus assumptions.
minimal assumptions are still just that, assumptions. especially if the data does not even support the minimal assumption.

simple logic can be simply wrong.

simple reasoning is still reasoning. still grey area. still assumptive.

the data, even as it is being assumed by you, does not back the claims you've made.

Upon peer review, the arguments you are making have been called into question, as the argument isnt adding up to your own assumptions.

so, with the lack of clear evidence/proof of your statements, that is merely an opinion.

your opinion. and you are entitled to it.

so maybe it should be prefaced that way;

'in my opinion, it is slower'.

please do not mislead our dear readers by stating opinion as fact ;)
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
No, I am making minimal assumptions.

Try to follow the simple logic. A claim was made that RAID-0 of two SSDs was always faster than a single larger SSD with TRIM.

I linked to data, and used simple reasoning, to show a counter example where RAID-0 of two SSDs would not be faster than a single SSD with TRIM.

Therefore, RAID-0 of two SSDs is NOT always faster than a single SSD with TRIM.

Please try to think about this and understand the basic logic. No need to post more nonsense arguments with bogus assumptions.
in light of the new data, I'm revising my claim from "always faster" to "worst case it matches performance but always faster when the solo trim drive is at least 25% full", but I don't even know if the matches performance is correct. they would have to run the test with a raid array of two drives that are 75% full against a single drive that is dirty/empty.

if someone asks me if they should get a raid array or stick with a solo drive for trim, I'm going to ask them if they plan to fill their drive to 25% of its capacity. and when they say yes, recommend the raid assuming they have a controller to handle it

Besides, the data makes no sense if there was no TRIM. After filling it up to 75%, how could the benchmark get much faster after deleting the files, unless there was TRIM or secure erase (and they would certainly not call a SE'd SSD "empty / dirty")? Just deleting files without TRIM or SE, the SSD does not know that the sectors have been freed -- the SSD would still have the pages corresponding to at least 75% of the LBAs marked as valid, despite the files being deleted. Unless they were TRIMmed, which they obviously were.
the same way the agility 3 is somehow faster at 75% full compared to 50% full... or how the corsair force 3 is faster after being marked dirty/empty than in its initial clean state.
 
Last edited:

Computurd

Gawd
Joined
Dec 19, 2008
Messages
956
yes!
funny how some of the drives are faster at 75 percent full than they are at 50 percent full. if i extrapolated the data and made a wild opinion out of it, i could say that they would be faster at 100 percent full than at 50 percent full.
how would that throw a wrinkle in the debate LOL?

seriously, the data can be interpreted any way you want, if you are comparing data that is unrelated to the argument that is being made.
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
By the way, note that in the counter-example I have been talking about, we are comparing a 75% full SSD to an empty/TRIMmed SSD of the same size. A single counter-example is enough to disprove "ALWAYS", so I think the matter of "always" -> "not always" can be laid to rest.

But it may be worthwhile to look at other cases.

To explore the difference between RAID-0 of two SSDs and an equivalent single SSD without RAID, we should compare the speed of an SSD that has been filled completely at least once with TRIM disabled, with an SSD of twice the size that has TRIM enabled. Note that the smaller SSD will have the flash pages corresponding to 100% of the LBAs marked as valid regardless of how full or empty the filesystem is on that SSD, since TRIM is disabled. In contrast, the larger SSD would have pages marked as valid only up to the percentage of LBAs that the filesystem is using, since TRIM would inform the SSD to mark the unused pages as invalid.

So, ideally, this is the test that we need:

SSD "P", capacity X vs. SSD "Q", capacity 2X

1) Prepare P: disable TRIM in the OS. Copy files to P until completely full. Delete the files.

2) Run benchmarks on P. Ideally, with the filesystem empty, and then partially filled in increments up to 95% (or whatever is the most that allows the benchmark to run)

3) Prepare Q: enable TRIM in the OS. Delete all files and run a manual TRIM (or secure erase) to ensure all pages are marked invalid.

4) Run benchmarks on Q, with the same procedure as used in (2).

tweaktown basically did steps (3) and (4) already. So ideally, someone needs to take the same model(s) of SSDs, but half the capacity, and run steps (1) and (2) on them to compare.

I am not aware of any published comparison tests such as described above. They may be out there, but I have not seen them.

If these tests were done, I guess that the performance in (4) would typically be greater than (2), but usually less than double the performance. I say "typically" and "usually" because it will depend on each SSD and on the specific benchmark used, as well as the fill amount in (4) [*]. So, the RAID-0 of two drives would usually be faster than a single larger drive with TRIM. However, the big question is how much faster, and whether it is worth the hassle of RAID and the increased chance of a failure with two SSDs.

It is my opinion that unless someone specifically needs very high performance, and probably only in the case of needing very high sequential performance, they will have a better experience with a single larger SSD with TRIM. Most people, even most power users, will simply not notice the difference in performance in their daily usage, so there is no point to adding the hassles of RAID and the increased chance of failures, unless there is a specific reason for going RAID.


[*] I expect (2) would have only small dependence on how full the filesystem is, in fact, it may be faster as the filesystem fills since the span of the benchmark test might be limited to a fraction of the available LBAs assuming the filesystem is not fragmented
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
the same way the agility 3 is somehow faster at 75% full compared to 50% full... or how the corsair force 3 is faster after being marked dirty/empty than in its initial clean state.
You are looking at differences of a few percent, which are in the noise. I am talking about differences of around 50%. Big difference.
 

john4200

[H]ard|Gawd
Joined
Oct 16, 2009
Messages
1,537
if someone asks me if they should get a raid array or stick with a solo drive for trim, I'm going to ask them if they plan to fill their drive to 25% of its capacity. and when they say yes, recommend the raid assuming they have a controller to handle it
I do not think that is the right conclusion to draw from the data and this discussion. See my other post for more detail. Basically, the data from tweaktown is not ideal for the comparison. What we really need is a 100% used (no TRIM) measurement of an SSD half the capacity. That will obviously be even slower than 75% full at same capacity. So you are likely overestimating the benefits of RAID by looking at the tweaktown data.
 

Computurd

Gawd
Joined
Dec 19, 2008
Messages
956
However, the big question is how much faster, and whether it is worth the hassle of RAID and the increased chance of a failure with two SSDs.
that is a valid argument, and definitely subjective to the individual user.

so.. who wants to run the test :)

i have tons of SSD, but they all are matching sizes.
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
You are looking at differences of a few percent, which are in the noise. I am talking about differences of around 50%. Big difference.
yeah, except they never dip below 50%. they float right on it for worst case. so even with their stuff it's possible the raided drives are still better, so unless you recreate the scenario where the raid is worse, so far it hasn't been documented. always still stands for putting any amount of data in that clean drive approaching 25%.

However, the big question is how much faster, and whether it is worth the hassle of RAID and the increased chance of a failure with two SSDs.
that's just subjective, I can't test if it's worth it for someone to do it. on some of the setups you save plenty of money too, like raiding c300s vs buying large.
 

gjs278

Gawd
Joined
Feb 22, 2011
Messages
711
so.. who wants to run the test :)

i have tons of SSD, but they all are matching sizes.
same... but I can offer a look at 2x array of 60gb force gt drives that have definitely been filled to the max and never trimmed.
 
Last edited:

Computurd

Gawd
Joined
Dec 19, 2008
Messages
956
ive got some absolutely ridiculous benchmarks. shit outta the movies LOL
all 8xR0 though. this is actually a lower one...
 
Top