SSD Peformance

[Dr J]

Hi, I just installed a new Samsung 250 gb SSD in my old laptop and I'm curious, does it work the same way as a mechanical hard drive, in that the more information you store on it, the slower it becomes?

thanks,

John

 

[kbp]

No, as long as you do not exceed about 70% capacity..

 

[Dr J]

So, having a secondary drive for data and digital pictures is unnecessary, as long as total volume is less than 70% capacity; there would be no performance gain by taking such an approach?

 

[Phynaz]

No, as long as you do not exceed about 70% capacity..

LOL, this number keeps getting bigger and and bigger.

Now we are up to leave 30% free. Next month somebody will be saying leave 50% free.

 

[Phynaz]

Hi, I just installed a new Samsung 250 gb SSD in my old laptop and I'm curious, does it work the same way as a mechanical hard drive, in that the more information you store on it, the slower it becomes?

thanks,

John

Why would you think a platter disk drive would become slower as you fill it?

As far as the SSD...You have it. Enjoy it. No special precautions are needed.

 

[Dr J]

What, then, is the reality, or is there no objective test for this?

 

[postmortemIA]

What, then, is the reality, or is there no objective test for this?

yes, there is anandtech article how some controllers suffer when there's not much spare area left:
http://www.anandtech.com/show/6489/playing-with-op

Hard drives have similar problem, outer sectors are much faster than sectors near center. Simply because there's much more data on outer sectors and disk head has constant rotation speed.

 

[Dr J]

postmortemIA, thank you!

 

[philipma1957]

Why would you think a platter disk drive would become slower as you fill it?

As far as the SSD...You have it. Enjoy it. No special precautions are needed.

you do realize that mechanical hdds really do slow when filled in?


as does an ssd.

the ssd will suffer if you have a big file to load on to it and very little free space. for instance a 120 gb ssd with 25 gb free will allow you to write an 18gb eyetv file more slowly then a 120gb ssd with 90gb free does. portmortemIA is correct

 

[Phynaz]

you do realize that mechanical hdds really do slow when filled in?


as does an ssd.

the ssd will suffer if you have a big file to load on to it and very little free space. for instance a 120 gb ssd with 25 gb free will allow you to write an 18gb eyetv file more slowly then a 120gb ssd with 90gb free does. portmortemIA is correct

Two things wrong with this post.

1. Please show how an HDD slows when filled in. Unless you are filled to the point the Windows can't defrag the drive it doesn't matter. And it certainly doesn't matter for photo storage.

2. SSD's only slow down when you are writing to them faster then garbage collection can do its job, which is not a typical scenario. Especially again for a storage drive.

In your case cited above, using just the info you have provided, both drives would write the file at the same speed. That's because you never stated that the drive wasn't given a chance to garbage collect.

Hope that clarifies things for you.

 

[Phynaz]

yes, there is anandtech article how some controllers suffer when there's not much spare area left:
http://www.anandtech.com/show/6489/playing-with-op

Hard drives have similar problem, outer sectors are much faster than sectors near center. Simply because there's much more data on outer sectors and disk head has constant rotation speed.

What Anand is doing is a torture test in which garbage collection isn't run. OP is operating a laptop on his drive. They aren't the same thing.

Why doesn't anybody get this?

One more time...A normal user isn't going to run into the situation of continuous non-stop writes to a drive. It's not a realistic situation. The over provisioning provided by the manufacturer is fine for normal usage.

Car analogy time...I can design a test that will blow up the engine in any car you can buy today by driving to work in first gear. Does that mean anything to a normal commuter? Just don't drive to work in first gear, which isn't something anybody would do anyway.

 

[Phynaz]

One additional point. Even if you do manage to torture a drive to the point that it slows down, it is only writes that slow down. Reads will be just as fast as ever.

And the slow down would be only temporary, until you gave the drive a rest.

 

[tracerbullet]

Why would you think a platter disk drive would become slower as you fill it?

I'll venture a guess that they are talking about data transfer from the outer edge of the platter being quicker than from towards the center. And quite possibly misunderstanding why and what that actually means.

 

[Phynaz]

I'll venture a guess that they are talking about data transfer from the outer edge of the platter being quicker than from towards the center. And quite possibly misunderstanding why and what that actually means.

Which would be a valid argument if our operating systems used disks in that way.

 

[Hellhammer]

1. Please show how an HDD slows when filled in. Unless you are filled to the point the Windows can't defrag the drive it doesn't matter. And it certainly doesn't matter for photo storage.

This comes from the fact that the outer edge of the platter has higher tangential velocity. It's just simple rotation physics, v=rw. w is angular velocity and it's a constant in hard drives, which is why only the radius affects the tangential speed (i.e. performance).

What Anand is doing is a torture test in which garbage collection isn't run. OP is operating a laptop on his drive. They aren't the same thing.

I don't see how the computer being a laptop has anything to do with this. You can run IO intensive workloads in a laptop as well.

One more time...A normal user isn't going to run into the situation of continuous non-stop writes to a drive. It's not a realistic situation. The over provisioning provided by the manufacturer is fine for normal usage.

The point is not the continuous writes, but filling the drive. If you fill 90% of the drive, especially random write performance will take a hit (most write IOs are random and the IO size is 4-8KB, so random write speed isn't just some synthetic figure).

Garbage collection is also more efficient when there's more OP. If 90% of the LBAs are filled, there is obviously less empty blocks available. Now, if you go and write something to the 90% filled drive, you will have even less empty blocks (eventually you may have none). Sure garbage collection will kick in at some point (some drives rely heavily on idle GC, whereas some are very aggressive) but it doesn't solve the problem in one second. It may take hours depending on the drive and the level of fragmentation.

It's not much to ask to keep some free space in an SSD. We have found 20-25% to be optimal from a performance standpoint, although this is again drive specific and e.g. Corsair Neutron is surprisingly consistent with the stock OP.

 

[kbp]

As I had said....ABOUT 30% free space is a very safe range, just in case you like driving in first.

 

[Phynaz]

This comes from the fact that the outer edge of the platter has higher tangential velocity. It's just simple rotation physics, v=rw. w is angular velocity and it's a constant in hard drives, which is why only the radius affects the tangential speed (i.e. performance).

Yes, we all know about platters having higher density on the outside of the disk.

But what you haven't proven is that Windows will fill the disk from the outside in, and causing some kind of user noticable performance drop as the disk fills.

 

[Phynaz]

I don't see how the computer being a laptop has anything to do with this. You can run IO intensive workloads in a laptop as well.


<snip>

It's not much to ask to keep some free space in an SSD. We have found 20-25% to be optimal from a performance standpoint, although this is again drive specific and e.g. Corsair Neutron is surprisingly consistent with the stock OP.

Have you published any test results showing any noticeable decline in SSD performance with a typical workload on a typical laptop SSD when filled to 90% of capacity?

Use cases matter.

You even wrote right in this forum you were having difficulty benching SSD's because they were all basically the same in real world performance.

 

[postmortemIA]

Yes, we all know about platters having higher density on the outside of the disk.

But what you haven't proven is that Windows will fill the disk from the outside in, and causing some kind of user noticable performance drop as the disk fills.

Right, it is so hard to get informed about this. Even wikipedia has nice articles about formatting and disk sectors. It nicely says there that a disk partition consists of consecutive disk sectors. That is only reasonable solution, anything else would yield in terrible sequential speeds for mechanical disks. If you do any disk benchmark against multiple partitions of the same disk, you'd see how 1st partition is way faster than the last one.

 

[tracerbullet]

If you do any disk benchmark against multiple partitions of the same disk, you'd see how 1st partition is way faster than the last one.

Yep, I have 4 partitions on my 3TB, the outermost benches about half again as fast as the inner one.

But what you haven't proven is that Windows will fill the disk from the outside in, and causing some kind of user noticable performance drop as the disk fills.

What are you looking for as "proof"? You think the hard drive manufacturers haven't figured this out or something? You could come up with lots of scenarios, all of which end up the same way - By placing more data, you fill the drive more, that data has to go onto an ever slowing region, and the overall average speed of reading or writing drop accordingly.

Why would you think a platter disk drive would become slower as you fill it?

Have you published any test results showing any noticeable decline in SSD performance with a typical workload on a typical laptop SSD when filled to 90% of capacity? You even wrote right in this forum you were having difficulty benching SSD's because they were all basically the same in real world performance.

SSD's aren't the same as platter drives. Why are you asking anything about SSD's? Isn't your big question / argument with how everybody thinks platters fill? Why are you bringing SSD's into this as some kind of comparison? Apples / oranges.

 

[Zap]

But what you haven't proven is that Windows will fill the disk from the outside in, and causing some kind of user noticable performance drop as the disk fills.

Windows (not talking about Defrag) doesn't know where the data is going, regardless of whether it is a HDD or SSD. It just tells the drive to write the data.

The drive will then fill from outside-in on its own. See this video. When off, head parks on the inside. After initializing, the head stays on the outside because it is ready to use. Then, it goes back to the inside when shut off.

 

[Phynaz]

I disagree Zap. The OS tells the drive where to place the data. That's the point of the file system.

All a drive knows about is its own sectors. It has no idea what a file is, so it wouldn't know how to write from the outside in.

 

[tracerbullet]

And you further contend what, that Windows just places data totally randomly and doesn't try to fill from the outside in?

 

[Hellhammer]

I disagree Zap. The OS tells the drive where to place the data. That's the point of the file system.

All a drive knows about is its own sectors. It has no idea what a file is, so it wouldn't know how to write from the outside in.

The OS doesn't have direct access to the platters or NAND, otherwise there would be no need for SSD or hard drive controllers (yes, hard drives do have controllers in them as well). Besides, SSDs/HDs are file system independent, they don't need a file system to work.

If you look at the graph I posted earlier, the performance goes down the more the drive is filled.

 

[Phynaz]

How does a HDD know if a section of disk is in use or not and can be written to?

It doesn't.

This is basic stuff, I can't believe people are failing on this stuff.