HDDs vs SSDs

To be clear - I am just saying I have never seen any sort of recommendation of this sort documented anywhere - and feel I have given it an honest search. If there is some white paper, a real review site, or SSD maker who makes a general statement about SSD drives saying x number of percent of the total disk space on SSDs should be left free for best performance, I would like to read it. I am here to learn too.

 

I've never heard an exact percentage given but I've always heard that a full SSD drive will be slower than one with just a fwe GB free.

http://en.wikipedia.org/wiki/Write_amplification#Over-provisioning

All modern drives should overprovision. I think my 830 has like 8GB or something free at all times.

I think that solves that question.

On a modern drive you have no need to keep free space - your drive has partitioned it off for itself. On some drives you may have to but it they don't support GC/TRIM it won't matter.

 

Excellent!

Even on a regular hard drive the rule of thumb for Windows has been to have a good amount of unused space for Windows to work efficiently. One example is Pagefile.sys which has a range of GB, minimum to maximum range based on the memory. If the hard drive is almost full and the Pagefile.sys is at the minimum range, then it has no empty space for Pagefile.sys to expand, thus resulting in decrease performance.

Best regards,

 

Great posts. Thanks guys.

 

I have read that in general NTFS partitions perform best with 30% or greater free space. I've seen a few claims of 10% but I find that pushing it a little bit.

 

Hi all

If it comes to cost here's my penny's worth.

I have 2 SATA3 6Gb/s SSD's, a very fast Corsair Force GT 120GB and a near as fast Crucial M4 256GB.

But the best is my Hybrid, A Seagate Momentus XT 750GB SATA3 6Gb/s @ two time the size & a third the cost of the other two together, it is no-brainer.

Take Care
TheQuest

 

Oh? Got links?

I don't see that the hard drive (or file system used) cares one bit how much free space is on it.

The operating system, on the other hand, cares very much how much free space is available.

 

If I come across the intel paper that talks about user-overprovisioning I will post it.

Regarding page files, it's just fine to put pagefile.sys on the disk. That's as long as you are not using 6 instances of photoshop on a 1GB machine. You should always have enough ram going. Today I doubt this is a problem.

 

So, the conclusion is that hard disk whether kind including SSD, need free space for Windows performance. This includes TRIM too!

 

The free space I'm talking about is to give the drive room for added overprovisioning.

Most of my newer SSD's don't slow down unless they are like close to 99% full AND doing lots of writing.

The purpose of allowing free space in an SSD is different than an HDD.
In an HDD the free space was to help with fragmentation. Go ahead and see how the HDD slows down when you're writing the last few percent and the free space is all over the place.

In SSD's the point of having breathing room is to allow the controller some room to prepare contiguous blocks of free space. And remember in SSD no matter how scattered a file is (typically hundreds of fragments) the speed is no slower. It all comes down to write performance.

Why not test it yourself? Fill the SSD and run some benchmarks? See what percentage your SSD chokes at when writing.

 

Before you posted all this, I have already posted this in my post #49 on page 2, please refer to it.

Best regards,

 

Huh? No! You cannot conclude that at all! At least I sure don't.

It all depends on what the disk is used for. If I jam my disk (I am not saying which type of disk) totally full of mp3 files - right up to the very last available byte of storage space, there is no reason Windows cannot access that disk and obtain the same performance from it as if it was only half full.

For "Windows Performance"? Yes, the "boot" disk (again, not saying which type of disk) needs to have some free space for the OS to store temporary files. The disk containing the PF file needs space for the PF to expand and contract (if not using fixed, and I recommend all Windows 7 users let Windows manage your PFs). And if you have moved your default Documents and Temporary Internet Files folder locations, those disks need space too.

You CAN conclude, however, when making purchasing decisions, you ALWAYS buy more than you will need - at least within your budget - to ensure the longest "future proofing" for your investment. That goes not just for disk space, but RAM, graphics horsepower, CPU horsepower, and last but certainly not least, your PSU.

Sadly, too many buy the biggest CPU they can find, but wimp out on disk space (or worse, don't buy quality power). I personally think swapping out CPUs is a lot easier than swapping out drives containing data (especially if the boot drive) so I always try to buy much more drive space than I think I will ever need - then double it.


*****

Not for trim (a command, not an acronym, btw). Over provisioning provides the space for that, and I note Windows 7 is very adept at ensuring garbage does not collect in the first place. Yeah, it does assume a recent SSD, but then we all have drawers and shelves of perfectly good, but obsolete drives (CD, DVD, HD, floppy, and now SSDs), cards, CRT monitors, network gear, and more, right?

Ummm, no, I am afraid that is not right at all. SSDs could care less about "contiguous" space. "Contiguous" space is a fragmentation issue. Your comment suggests the controller (which would be the "SATA" controller) is defragmenting the free space in the background. Defragging the free space requires moving reading and writing into different locations) constantly!!! "fragments" of data. That is not happening unless I am totally wrong - would not be the first time. Got a link 'cause I have never seen that documented anywhere, or heard of that before and would need to read up on it.

We know that Windows queries all drives for type during boot, then disables defragging on the SSDs. My bud Bing Google so pretty sure I am right on this.

And note,

"Write" performance? Benchmarks? I guess if write performance and benchmarks are that important to you, then you have a point. But disk reads BY FAR way outnumber disk writes, even on file servers, so if write performance on a SSD is impacted that much to affect the user, then there are several other bottlenecks (besides too small of a disk) on that system that need to be addressed. If shooting for bragging rights (and nothing wrong with that - if that is one of your objectives) then go for every advantage available - even the tiniest add up.

Here are my conclusions:

• If disk space is a concern, you bought too little. Buy more.
  
• There is NO supporting evidence (from the makers or any professional review site) to support ANY claims that x-percent of free space is required for SSDs. A full disk, (regardless of type) WILL have write performance issues. Therefore, "some" free space is aways desired. Remember, when prompted with "[filename] already exists. Do you want to replace it?" does not actually "over-write" the data in those file locations. The new save has to be verified before the old space becomes available again.
  
• The "boot" drive (regardless of type) needs enough free space for the OS and PF to operate in. However, there is no "standard" recommended percentage. 10% of 40Gb is not enough, 10% of a 2Tb drive is way too much. My experience says 10Gb of free disk space is not enough. 30Gb or more free space on the boot drive is what I recommend - but even then, periodic crud cleaning is essential to keep the free space free.

To be sure, I am not suggesting anyone fill your drives to the max. I totally agree that as a general statement, plenty of free disk space is good, and a priority when planning your next PC or upgrades. My issues were with claims that SSDs required 20 or 25% free space to keep performance up and no supporting links for those numbers.

 

An SSD can really only write at the block level. But it can read at the sector level. When writing or updating one byte. It must either find space(blank blocks) someplace, or erase a whole block, and in the process of erasing a block it will copy a bunch of unrelated data(to the file you're working with) from one block to the other. And perhaps erasing and even moving more.

When an SSD is full, this can cascade out of control. And you'll see a slowdown.

Overprovisioning gives the controller room to execute TRIM and IGC which prepares contiguous sectors that are empty. Contiguous empty sectors within a block means no block erase/copy operations are needed. Newer SSD's are doing this more and more in the background.

Understand that blocks may be all over the place and going from one to the other incurs no speed penalty. But copy erase ops do!

Please read up on Write Amplification.

 

That's good information but I don't think any of that was in dispute - at least not by me. Since blocks are typically no bigger than 512Kb, the cascading event you describe would not occur unless free space was at a critical level.

That's too late. Something should have been done long ago before it got to that point.

Okay. Old stuff so not sure your point.

I think there's still some confusion over contiguous. If something is contiguous, it is not fragmented. And when it comes to blocks on an SSD, that WA is only an issue due to how erase only works on the full 512Kb. But again, that's where Trim, Windows 7, and current drives come in.

 

Bill,

I have trouble accepting that the more RAM you have the bigger the pagefile needs to be. Say you have 20 GB RAM. The Microsoft recommended pagefile is 30 GB. That's a big chunk of a SSD.

This is an old link and it's written for WinXP. I use the recommendations for my WinXP OS. 400 MB pagefile. I've never seen my pagefile use more than 200 MB. I'm currently using 26 MB.

http://aumha.org/win5/a/xpvm.htm

I don't have any similar links to pagefile size for Win7.

 

Windows XP is a different breed than Windows 7. Usually, it is recommended to turn off the Pagefile on a SSD. If one does that or one has less than 2GB Pagefile with lots of memory, one will have real erratic behavior on boot up and the mouse pointer will be running everywhere. After boot up things will be stabilized.

For Windows 7, whether one has SSD or not, it is better to leave Pagefile as recomended by Microsoft, which is higher the memory the higher the Pagefile. Or, at least a minimum and a maximum of 8GB.

Best regards,

 

I'll go with this one from Microsoft that states 20% or greater free space and leave it at that:
http://support.microsoft.com/kb/959062

I didn't claim the hardware cared. NTFS Is part of the operating system, namely Windows, which is what 90+% of people will be currently using. FAT variations of file systems will prefer different amounts, which I am not going to bother to research for the sake of argument. Storage only disks with static files will of course need little to no free space, but that did no appear to me to be the point of the discussion.

 

I found the following numbers interesting and I wonder how they fit into the "free space" discussion.

My Win7 32 bit has a total partition size of 18.5 GB with 3.20 GB of Free Space. The Win7 Index reports a Disk Data Transfer rate of 7.8.

My Win7 64 bit has a total partition size of 11.2 GB with 1.00 GB of Free Space. The Win7 Index reports a Disk Data Transfer rate of 5.9.

I resized the Win7 64 bit partition to a total size of 12.6 GB with 2.47 GB of Free Space. The Win7 Index now reports a Disk Data Transfer rate of 7.7.

So increasing the Free Space made a difference to this figure. Is this meaningful?

 

Dear Brian,

Long before all this conversation started, I told you in my earlier post on page 2 that you need 25% to 33% for Windows performance and TRIM. One example I gave is that Pagefile.sys has a range from minimum to maximum and needs space to grow.

Best regards,

 

aladdin,

There are two views in this thread. One that you need 25% of free space on the SSD and it doesn't matter if this free space is inside or outside a partition. The other view is that you need negligible free space on a SSD. But does the second view imply that an OS partition can be "almost" full of data? Bill, am I misquoting you?

 

Concerning the pagefile, Mark Russinovich makes these comments....

"There’s no end of ridiculous advice out on the web and in the newsstand magazines that cover Windows, and even Microsoft has published misleading recommendations."

And concerning letting Windows manage the pagefile,

"whoever wrote that code got their guidance from one of those magazines I mentioned!"

His recommendation for pagefile size is along the lines of the more RAM the smaller the pagefile. A lot of RAM would allow a very small pagefile. But read his blog.

http://blogs.technet.com/b/markrussinovich/archive/2008/11/17/3155406.aspx

 

The only reason to use page/swap is if at some point your system will have to put pages into it.

If your system never has to do this you'll never need the pagefile.

EVen if you never *need* the pagefile a program may still use it. That's why people turn it off on Windows.

On Linux you can just turn swappiness down to 10 or you can disable it altogether.

If you turn your pagefile up you won't get dumps from kernel panic.

 

Dear Brian,

If you have Windows 7, lots of memory and SSD. Then try a small Pagefile.sys and a large Pagefile.sys

You will find that with small Pagefile.sys, you will have very erratic behavior on boot up. Freezing and your mouse cursor jumping uncontrollably every where. After a while it will stabilize. This happens only on booting up.

Do your own experiment?

Best regards,

 

aladdin,

I just tried Win7 (SSD) with a 16 MB pagefile (20 MB maximum). No problems at all.

 

I do not know about ssds but from various test, that I run windows OS will perform fine with as little as 2 GB free on the system partition. (

My guess is that the guidelines of having 25% free where about either ancient disk capacities and/or about the need to defrag them efficiently (which on ssds is not an issue).

Panagiotis