Using SSDs with Windows XP

[o_o]

Hi,

I am looking at using a SSD, around 240 or 256 GB with Windows XP Pro. I have no experience using SSDs, so I am clueless about some of the technical terms thrown around.

Searching for information on this topic reveals a number of old threads that suggest SSDs should not be used on Windows XP due to lack of TRIM support. However it seems that newer SSDs have improved Garbage Collection, so may work with Windows XP. Is this true?

In this regard, which would be a good SSD for Windows XP? I have heard that the Samsung SSDs are better than the ones using the Sandforce controller based SSDs, due to better Garbage Collection features.

Which SSD vendor offers software to perform a manual trim on Windows XP? Or is this even needed on Windows XP? (This may technically be called running manual GC, I am not sure.)

For the new SSDs, do we still need to align the partition before installing Windows XP? Also would Windows XP recognize, and install itself on the SSD, or do I need specific drivers? (I have a Win XP SP2 install CD.)

Thank you in advance for your help.
O. O.

PS: Please, no discussion on Why I am installing Windows XP, and that it is a bad idea etc. I understand, but that is not my question here.

 

[Cerb]

Searching for information on this topic reveals a number of old threads that suggest SSDs should not be used on Windows XP due to lack of TRIM support. However it seems that newer SSDs have improved Garbage Collection, so may work with Windows XP. Is this true?

Yes. Things are generally much better, these days. Not to the point of that being a 100% solved problem, but nothing like up through '10.

In this regard, which would be a good SSD for Windows XP?

Many, but Sandisk's Ultra Plus, Crucial's M500, Plextor's M5S, and Corsair's Neutron, will be good choices w/o TRIM, and are fairly cheap.

Which SSD vendor offers software to perform a manual trim on Windows XP? Or is this even needed on Windows XP? (This may technically be called running manual GC, I am not sure.)

Samsung and Intel, at the least, but your best bet is to just add some over-provisioning to an already-good drive, and not worry about it. When making the partition, make it several GBs less than the maximum size.

SSDs use more space than they give you to help handle wear-leveling (usually, the difference between GiB and GB). The more space they have to work with that isn't being used, the better performance can be. TRIM allows adding to that, by telling the SSD what space you aren't using, because otherwise it won't know (deleting a file normally does not clear out the space it used).

Pretty much all of them allow a trick to increase the amount they can use as spare, beyond what is set at the factory, by leaving unpartitioned space at the end of the drive. So, for example, taking a 256GB drive (about 240GB usable), and making an aligned 220-230GB partition, at the beginning (but aligned), would give it some extra spare room, improving performance over time, without worrying about TRIM. Get a drive with good steady-state performance from the outset, like the listed ones, then give it some extra GBs to work with right off, and you won't need to fuss around with manual TRIM runs, to keep performance acceptable over time.

For the new SSDs, do we still need to align the partition before installing Windows XP? Also would Windows XP recognize, and install itself on the SSD, or do I need specific drivers? (I have a Win XP SP2 install CD.)

Yes, it still needs to be aligned, for best performance and wear, if installing anything older than Windows 7. XP will see it as a regular HDD. You may or may not need SATA controller drivers, but that aspect will be no different than if using a HDD. Be sure to disable defragging.

 

[Deders]

Samsung's drives are great and they offer ways to optimise and clean the disk from within their utility program.

 

[It's Not Lupus]

Samsung's drives are great and they offer ways to optimise and clean the disk from within their utility program.

yes. intel too.

 

[Deders]

yes. intel too.

I believe Intel's newer drives can't compete with most other drives in terms of write performance (according to the graphs in front of me). Most other drives are more than 2.5x faster at incompressible 4K writes alone. Factor in that the latest Samsung drives have will last 11 years and it's a no brainer.

 

[John Connor]

What's this over provisioning about? When you have a Sand Force based drive it in its self consumes more space than a non Sand force drive. So I'm thinking out of the box it really is already over provisioned.

OP, I have XP Pro 64 on a SSD. Make sure you use SSD tweaker and do not defrag a SSD. Also, if you have a platter in the computer in addition to the SSD put your temp directory on the platter.

 

[Cerb]

What's this over provisioning about?

Flash wear-leveling, independent of the file system, needs extra space to write to. Generally, the more of that space there is, the faster the drive will operate, because it will have more unprogrammed space to play with (often, the only major difference between a consumer and enterprise drive is the factory over-provisioning amount). With TRIM, that space is equal to the factory OP plus free space on the file system (with some lag time). Without TRIM, just the factory OP. To work around that, most SSDs are clever enough to read an MBR partition table, and if there's blank space at the end, assume that it was left there for it to use. A handful may use any unpartitioned space, but space at the end being treated that way is pretty universal, now.

Some drives have used more spare space for themselves, like the Intel X2 and 320, 1st-run Corsair Neutrons, and Seagate 600s, primarily to keep performance high, without also increasing the wear on the flash unnecessarily. The partitioning trick allows you to do that. With a TRIM-enabled OS being the common case, now, and with user-added OP as a common option, they've trended towards keeping the GB v. GiB difference as their minimum OP. Without the user-added OP trick, that would be generally bad, but with it, it's a fair compromise.

http://www.storagereview.com/intel_ssd_520_enterprise_review
Take the now-canonical 520 article: 20% offers immense improvements in performance over time. Typically another 5-10% is plenty. The Intel 710, gains less, but it comes with 37% OP from the factory, rather than 7%. AT's reviews tend to show differences with user-added OP, too (generally on the consistency page).

When you have a Sand Force based drive it in its self consumes more space than a non Sand force drive.

Not so. Most SF drives have the same OP as everybody else, about 7.3%, or (1024^3-1000^3)/1000^3. The rest of the space is taken up by an added error-correction enhancement scheme, RAISE. There are now SF drives with RAISE disabled, giving you the same space as anyone else's SSD. 1st-gen Corsair Neutrons, and some current Seagate 600s, also take more space from you, for performance, while the Samsung 840 does it mostly to keep from high WA killing the TLC flash, for users that keep their drives fairly full.

 

[Deders]

Interesting, do you thing 20% would benefit all drives or just Intel/Sandforce based ones?

 

[Cerb]

Interesting, do you thing 20% would benefit all drives or just Intel/Sandforce based ones?

Example:
http://www.anandtech.com/show/7006/sandisk-extreme-ii-review-480gb/2
The Samsung 840 Pro, FI, shapes right up with more OP...it's just expensive for what it is, right now.

Edit: this one is probably a more on-topic comparison:
http://www.anandtech.com/show/6553/sandisk-ultra-plus-ssd-review-256gb/4

Pretty much all of them, to varying degrees. Though, 20% is probably a bit much. They are running stress-testing software, which is an attempt to show the drive at its worst. Recent drives that have prioritized consistency (Intel 320, Corsair Neutron 1st-gen, Sandisk Extreme II, Seagate 600...) have generally taken ~12% total, which would amount to <10% from user-available capacity of drives with power-of-two sizes. The performance will vary more by workload and actual space, than relative space v. capacity. IE, 10GB spare in a 500GB drive will give you about the same as 10GB spare in a 100GB drive, for the same data being written. So, it's not like you need 20%, they just tested it with values high like that to show the difference. The lesser middle-ground used by other drives to get 60, 120, 240, etc. capacities, is probably a good target, and will still leave you with most of the drive's space to put data in.

But, since their sizes were also for marketing reasons, since people don't want to be thinking about a 245GB SSD, being a little off in either direction won't make much difference. On a 256GB drive, already shown to offer pretty good performance consistency, give it another 10-15GB, so it's in the ballpark of a 240GB, and then just don't worry about it, once the OS is installed and set up.

Also, if you get a drive with software, like a Samsung, Corsair, or Intel, you can use that software to check/apply added OP (I know Samsung and Corsair's will), as a way to know it's being used for that.

All that said, if you just want something that'll stay way faster than a HDD, any SSD with good steady-state performance will be fine with little more than partition alignment, for light/medium use, just not as fast over time as with TRIM or more OP space.

 

[o_o]

Thank you very much guys: Cerb, Deders, It's Not Lupus, and John Connor for your explanation. I am sorry I could get back to you earlier.

Thank you Cerb for the detailed information. Can this software that comes with the SSD be used before we install the OS?

From your description I think I should first create a partition that leaves a bit of space towards end of the drive e.g. if 240 GB is total space, set it to 220GB. Next, install the OS. Next run the SSD software to tell the SSD what the Over Provisioning space is. Is this correct?

Maybe I am wrong here, but it seems that Over Provisioning certainly improves performance. However I am wondering if Over Provisioning helps improve the life of the drive, by wear leveling or some other mechanism?

Thank you again,
O. O.

 

[Cerb]

Maybe I am wrong here, but it seems that Over Provisioning certainly improves performance. However I am wondering if Over Provisioning helps improve the life of the drive, by wear leveling or some other mechanism?

Yes, but the difference is going to be generally negligible, for most users, on account of (a) how much added OP helps depends on how much of what kind of data is being written over time, and (b) that SSDs have trended towards reducing WA over keeping performance really high, though how far they go varies (average performance is much higher, though). Basically, it's way better, on average, today, than it was even 2-3 years ago.

FI, if you have about 18GB spare (typical 256GB drive), and it takes you days to fill that up, mostly with writes 64KB or larger (lazy writes from Windows' IO buffer(s), mostly), the drive will have plenty of time to ideally handle the wear, though a little more space might pep it up, some. Some drives, like the Samsung 840 (TLC non-Pro), act more or less paranoid about it, and slow down a good bit (though still not down to near-HDD speeds, like some older SSDs), but maintain low WA (TLC flash can't take as many writes as MLC, so that was not a bad decision on Samsung's part). Others act similarly, but to a lesser degree, in that they tend to slow down, and perform wear leveling as well as is possible with the space they have, rather than let WA get too high, unless you have a workload that's really suited more to an enterprise drive. Sure, you'll have some high WA during short IO-intensive periods, but it averages out.

Some older fast JMicron and Indilnix drives, such as the Patriot Torqx, had write amplifications that could exceed 100x, while most SSDs today, even that do get higher WA by favoring higher performance (not that writing more means more performance, but that optimal moving and writing for low WA might take longer to complete a given write, compared to just writing wherever happens to be free, right now), tend to average closer to 2x, and with light users, 1.05 to 1.5x.

You would need a couple decades to wear out most drives, even with moderately heavy workstation work, and several times that for a typical web/mail/game machine. So, some other failure will happen, first. Lots of tiny random writes can wear them out much quicker, since the effective overhead is the same whether it's 4K or 128K, but the larger writes will so dwarf the small writes that it's just not an issue, today, 99% of the time. If it's almost all 4K or 8K writes, and lots of them (IE, DB servers), you can wear them out somewhat quickly, sure, but that simply will not happen, except in servers. Performance w/o TRIM can still be variable enough, but you're just not going to wear a current SSD out, even a TLC one.

 

[John Connor]

I'm not understanding. I have a 128 GB Adata in Windows XP Pro 64. It's damn near full and I have been installing stuff on the platter instead. I can't remember the amount of space I have left, but i'm thinking of leaving it as is for OP. I think the space is around 15 GB. So let me get this straight. You don't have to use software for OP? I can just leave the unused space as is?

I'm on my laptop right now and when I get on my desktop I'll update.

 

[Cerb]

I'm not understanding. I have a 128 GB Adata in Windows XP Pro 64. It's damn near full and I have been installing stuff on the platter instead. I can't remember the amount of space I have left, but i'm thinking of leaving it as is for OP. I think the space is around 15 GB. So let me get this straight. You don't have to use software for OP? I can just leave the unused space as is?

In Windows 7+, you can just leave empty space in your FS. That's why automatic TRIM support is so nice: free FS space = additional spare flash. With Vista or older, you either have to TRIM with special utilities periodically, or give it a few more GBs, right off, for the same kind of benefit.

It goes like this:
1. You get a brand new drive, and it should come blank, in a secure erased state, with all the flash unprogrammed, and no LBAs written to. Starting from this state is key, if using an older OS.
2. Some amount of the space (usually 7.3%, but there are exceptions) has been set aside for wear-leveling, and you can't go lower.
3. Making a partition that leaves empty space also leaves a bunch of LBAs that will not be written to. So if you made a 100GB partition in a SE-state 128GB (118GB) drive, you'd be leaving the SSD about 18 more GB for spare area. The LBAs in those 18GBs will stay in an unwritten state.

In Windows 7 or 8, you can do the same after the fact by making a partition, formatting it, and deleting it (what AT does for its reviews, I believe), too.

Exactly how they implement it can vary a bit, but almost all of them will work with the above. I'm not sure the history of it, and it may have started somewhat accidentally, but now it's a common, and often expected, feature. I can't find which, now, but I recall at least one at some point checked the partition table, but having a used/unused bitmap (mapped by LBA or some fresh flash block layout) would be better way to go, and AFAIK, is how most of them do it, now.

 

[o_o]

Thank you very much Cerb for that detailed explanation. You seem to really know a lot. I am sorry for the delay in responding.

Thank you for explaining how Over Provisioning reduces Write Amplification and so increases the life of the SSD. I always thought that writes and erases operated on the same sized memory, so I did not understand how OP reduced WA. But this is wrong. Writes always operate on pages which are 4KB, whereas erases occur on blocks which are 256KB. So Over Provisioning helps keep at least some blocks free to write to, so the Garbage Collector does not need to make multiple adjustments even when the drive has reached its total capacity (or you don’t have TRIM like WinXP) to find contiguous space.

Thanks again Cerb.
O. O.

 

[Cerb]

Writes always operate on pages which are 4KB, whereas erases occur on blocks which are 256KB.

Not quite. Both vary. 8K is increasingly common, and some are up to 16KB. However, the ATA standard uses 512B or 4KB for logical addressing. Windows, FI, doesn't get to know what the flash page size is (nor should it, really, without a file system designed just for flash memory).

So Over Provisioning helps keep at least some blocks free to write to, so the Garbage Collector does not need to make multiple adjustments even when the drive has reached its total capacity (or you don&#8217;t have TRIM like WinXP) to find contiguous space.

It does have to make adjustments when it reaches capacity, but the more space over-provisioned, the easier a time it will have. All available SSDs set at least 7.31% aside (that's 18 gigabytes for 256GB). For any given workload, the returns from over-provisioning will reach a point of greatly diminished returns, like a f=1/x graph, and with larger drives, you might not even need as much as it comes with form the factory to reach that point. Most available info about exactly how much is for whatever use is either old (SSD makers are keeping their research data private), or obfuscated by marketing.

Look at how it works like this. You've got the spaces A-Z to work with, each with limited endurance:

ABCDEFGHIJKLMNOPQRSTUVWXYZ

If you write to all of them, and are 100% full, then writing to any one left means wearing out that one. So if B stored important file system metadata, and was written to a lot, you would have a failure in short order, even though other letters may have had thousands of writes left.

So, what if you remapped it? You could have A-Z show up, but an a-z under the hood that might not match up. So writing to B, you would...well, where else can you write it to? :hmm: What if you had more space?

ABCD EFGH IJKL MNOP QRST UVWX YZ
abcd efgh ijkl mnop qrst uvwx yz12 3456 7890

So now that write to B can go to 1, instead of b. A request for what's in B now reads what's in 1. Also, b can be freed for later use. So, you then write to I, J, O, T, W, and Y. Now there's 6 freeable spaces.

But, there's a catch! Every 4 characters must be erased together! So a-d, e-h, i-l, m-p, q-t, u-x, y-2, 3-6, and 7-0, must be erased in their groups. From the above, with b, i, j, o, t, w, and y discarded, and now all space written to, to write more, some groups of discarded data need to be erased. To do that, some resting data must be copied, to preserve it. Instead of waiting until all the spare characters are used up, once there's been enough discarded data, you'd want to go ahead and start making some free room. So, one might copy k and l to 7 and 8, and then erase ijkl. Once ijkl are written to again, by that copying, 50% of a write was used on top of the host's write request, thus giving a WA of 1.5.

Extend that out to millions of such characters, and hundreds of them per erase group (well, 64-256 seem to be common), and you've got flash drives that perform wear-leveling. On the bright side, with most users, though, it turns out that lots of files written are quickly deleted, and that even without TRIM, that results in the same logical addresses being used over and over again (temp files, browser cache, auto-saved document backup files, etc.), and that a lot of writing gets done with sizes that are pretty large, so that fine-grained slicing up of pages, like above, isn't needed too often. Thanks to that, actual WA for desktop and notebook users can be kept moderately low, with or without TRIM. Exactly what to copy, where to copy it to, and when to copy, varies by drive.

In practice, lifespan problems only exist to very heavy random-writing users, and still project out to years. Adding a little more OP for performance over time w/o TRIM can make sense, since the performance drop over time can often be perceived, if the writing you do exceeds whatever the SSD can lazily do with its spare area (which TRIM effectively extends into your file system's free space). But lifespan issues have pretty much been taken care of (the people for whom it might not be generally know it).

 

[o_o]

Thank you for the detailed information Cerb. At least now I know I don’t have to care about SSD lifespan. I plan to give it about 20 GB for OP on a 256 GB SSD i.e. about 8 percent. I think that would be sufficient for me, since this Win XP machine would not be a server or something like that.

O.O.

 

[ashetos]

To work around that, most SSDs are clever enough to read an MBR partition table, and if there's blank space at the end, assume that it was left there for it to use. A handful may use any unpartitioned space, but space at the end being treated that way is pretty universal, now.

This sounds really scary, does this actually happen? Do you know some models that are partition aware at the firmware level (I don't care about software)?

So if I decide to hide a partition, change partition order and stuff like that, do I risk trimming the whole partition?

 

[Cerb]

This sounds really scary, does this actually happen? Do you know some models that are partition aware at the firmware level (I don't care about software)?

So if I decide to hide a partition, change partition order and stuff like that, do I risk trimming the whole partition?

I recall one controller being seen doing that, a few years ago, but I can't find it via searching right now. I doubt anybody would do that, now, though, between it being stupid (kind of like 512B at a time programming) and unnecessary.

 

[VirtualLarry]

but you're just not going to wear a current SSD out, even a TLC one.

Well, unless you do DC.

 

[johny12]

Well, I dont think just numbers can decide a product. I am not saying they are not important, they are but at the same time stability & actual performance matter a lot to me. I have heard there has been a lot of wrong perceptions about sandforce based drives, which I belive is getting wiped out forever. Give these drives try on your own to start believing!

 

[John Connor]

I have two SSDs that use Sandforce. I personally haven't had any troubles.