Z Series SSD question and RAID/TRIM concerns

Hello,

Does anybody know if the SSDs on the Z series use the slower MLC flash memory or do they use the more efficient (and expensive) SLC flash memory?

Thanks

 

MLC

Consumers aren't willing to pay the premium for SLC or 15k rpm drives, yet seem more than willing to pay twice as much for a CPU that can run 15% faster.
I'll never understand them.

 

MLC? Wow, that sucks!

Some of those Z series laptops can hit 4K in price, yet they skimp on the quality of SSDs they put in there. One might as well go back to a mechanical HD if it were an option, but on the Z series, it ain't...

And I agree with you. I'll never understand that, either. Not at that price.

 

The gap between MLC and SLC has closed dramatically. In fact, its rumored that Intel's next gen X25-E will use MLC chips. The controller is by far the biggest contributor to performance and reliability.

 

That's still rumors. I'm talking about right now.

SSDs still have a couple of more years to go before I'll consider them an option.

 

I just wondering what would i do if i brought a new vaio z and the ssd broke down...
sony customized sata slot and there seems not a lot of space for regular ssd,if that happen,half new vaio z price for repair?

 

You are missing out. SSDs are the single biggest upgrade you could do for your computer. Its really a whole new world.

 

Peteman is right - I have a Z91JS with 256gb SSD and its fast. Unbelievably the new quad ssds are way faster still. The Z ssd's are good, reliable and fast imo.

 

I'm sure they will be, but for right now for the reasons I mentioned above, I'll wait.

 

I hope my Vaio Z pre-order is worth it. I went for 192GB, which means 3 drives 64 GB each. Is it possible to set up so that I can use two 64GB drives for RAID1 and the third as an extra for various storage (which means 128GB in total)?

If I can, do operating systems perform a lot of writes and erases during normal tasks like internet browsing? I just want to know whether I should place the system into RAID1 or the third drive.

 

RAID 1 for the system drive, definitely. With RAID 1, writes are about the same speed as for a single drive, because the writes occur in parallel. Reads, on the other hand, are up to twice as fast -- exactly as for RAID 0.

Also, RAID 1 will allow you to run a wipe (e.g. with HDDerase) of one of the drives, and then rebuild, do it on the second drive, and rebuild again, without losing any data or having to restore from a backup.

In theory, TRIM should also work with RAID 1 drives, but unfortunately, Windows 7 doesn't pass on the TRIM command.

 

I also use a 48gb Gmonster ssd express port card for storing my files and games on which works really well and boosts my total storage capacity.

With a 32gb SD card and 16gb Sony memory stick duo it takes me up to 352gb of memory.

 

But also, I don't want to see a lot of write/erase tasks, since that would deteriorate the SSD over time if there's no TRIM. Say if I want to boot up the system, or let's say browse the internet, or even open up a media player; even if I don't save anything or record anything, will there be a lot of writes and erases?

 

Yes. You can check this now on any Vista/W7 machine, by opening up the task manager, then on the Task tab, go to "Resource Monitor". The "Disk" tab will give you a rough overview.

That said, you can reduce this quite a lot by moving your user home folder to a different drive. Then temporary internet files, icons, user registry and similar won't use the system drive.

And, if you have RAID 1, at least you can reset each drive to a non-saturated state without losing any data or having to reinstall anything. Lack of TRIM becomes less of an issue that way, if you have the time to preemptively do the manual operation every now and then.

 

Good grief, if you're going to do RAID setups then buy a second drive. Get rid of the optical drive and install a second drive in it's place and make sure your paging file is pointing to the second drive.

LOL.. Further reason to stick with mechanical drives.

 

I was thinking buying the new VAIO Z but honestly, I don’t know anything about SSD. What should I know before buying? Does it require a lot of maintenance?

 

Hi, Arth1,

Sorry, but your posts regarding RAID are quite misleading. First, the fact that reads and writes are "twice as fast" in RAID1 ignores the fact that the exact same data is being read or written twice - simultaneously. Why not ask for ten drives to be written simultaneously? The user will see no throughput difference. RAID1 SSD arrays are one-half the speed (in user throughput) of RAID0 arrays. Sorry!

It is true that RAID1 can make some sense with rotating drives that more frequently fail, but to throw away 50% of your storage on SSD drives which virtually never fail is by no means an obvious conclusion, and mine is quite the opposite, given the transformation in MTBF.

Second, your earlier posts which talk about not being able to do a restore with Acronis, and regarding splitting RAID arrays to do an HDDERASE, in addition to your point here about TRIM and RAID1, are equally off-base. The Intel controllers aren't offering you soft RAID - they are specifically offering a H/W- plus driver-based RAID solution. Acronis cannot tell if you are running RAID or not, nor can Windows 7. You absolutely can do a perfect restore with Acronis (including sector-based, if you really love wasting storage space). It is in no way necessary to reinstall Windows after running HDDERASE. Just reconfigure the RAID array in the BIOS and boot up the Acronis restore disk. Acronis trivially restores the MBB, all partition data (including resizing to different drives), and all disk data in a single step. It also is NOT Windows 7 which doesn't pass TRIM commands to RAID arrays. It's that RAID drivers do not currently publicize the ability in their code to accept TRIM. If they do, Windows 7 will simply pass them along. Once the authors of these drivers get their act together (which is in fact rather complex with many drive/RAID combinations), then TRIM will work with RAIDs.

Regards,
Mark

 

For writes, yes, it writes the same data in parallel, as I already said. For reads, it works exactly like RAID 0 -- if a file is two blocks in size, it can read one block from one drive at the same time as it reads the second block from the second drive. Think about it for a moment.

Single drive: A B C D E F G H

RAID 0 drive 1: A C E G
RAID 0 drive 2: B D F H

If you want to read A-B-C-D, you send two reads to each drive. A and B is read simultaneously, then C and D are read simultaneously. With me?
Now let's consider a RAID 1:

RAID 1 drive 1: A B C D E F G H
RAID 1 drive 2: A B C D E F G H

If you want to read A-B-C-D, you send two reads to each drive, just like for RAID 0. There's no advantage to reading the same data from more than one place, so the RAID doesn't. It uses the exact same read sequence as it would for a RAID 0: A and B simultaneously from both drives, then C and D simultaneously from both drives.

For mechanical drives, it will be slightly slower than RAID 0, due to more head movements, but seldom enough to measure. For SSDs, there's absolutely no speed disadvantage for reads for RAID 1 compared to RAID 0. Again, think about it!

Out of curiousity, who misinformed you so?
Again, sustained write speeds are slower, yes, and I've never said otherwise, but read speeds (and tiny writes) are the same speed.

For a read speed advantage without greatly increasing the MTBF -- actually reducing it. A four drive stripe is sheer folly, regardless of media. See one of my other posts for the maths. You only do that if you know you can afford to lose everything, and the speed advantage makes it worth it.

Again, you're very misinformed. The name of the RAID system is "Intel Matrix Software RAID". Guess what the "software" means?

What the controller gives you is an interface for writing RAID markers to the drives, so the software can assemble them. And it allows non-DMA reads of the drives through the BIOS firmware to allow the boot loader to load -- at which point the software drivers take over, assemble the RAID, and read/write it through special drivers that does the striping/mirroring purely through software.

Don't believe me? I have an Intel ICH raid on my server over in the other room. Look what Linux tells me:

Code:

# dmraid -r
/dev/sdd: isw, "isw_ijbcgfjei", GROUP, ok, 1953525165 sectors, data@ 0
/dev/sdc: isw, "isw_ijbcgfjei", GROUP, ok, 1953525165 sectors, data@ 0
/dev/sda: isw, "isw_ijbcgfjei", GROUP, ok, 1953525165 sectors, data@ 0
/dev/sdb: isw, "isw_ijbcgfjei", GROUP, ok, 1953525165 sectors, data@ 0

# dmraid -s
*** Group superset isw_ijbcgfjei
--> Superset
name   : isw_ijbcgfjei_Volume0
size   : 3907038976
stride : 128
type   : raid01
status : ok
subsets: 2
devs   : 4
spares : 0

Notice how each drive is available with device name? With a hardware RAID, this would not be the case -- the OS would only see a single drive.

If it were a hardware RAID, this would be true. But it isn't.
When the Acronis restore media boots, it sees the individual drives, just like my above linux system does. Unless it has drivers that can identify and assemble the RAID based on the markers (written near the end of each drive), it can't do anything. And if you have a broken RAID (like a set of new drives that haven't been set up yet), the markers won't even be there.
The last time I tested Acronis, it did backups just fine. It just couldn't restore them, because the very limited boot environment didn't have drivers.

Sorry, you're wrong again. If you have access to MSDN, you can find information on how TRIM is implemented in Windows. It's part of the new Microsoft ATA driver -- Windows 7 asks the ATA driver whether the drive has the "supports TRIM" flag set, and only then will Windows start sending TRIM commands. This, unfortunately, means that if you use anything except Windows ATA drivers, it won't be seen by Windows, because Windows only queries its own drivers for the flag.

It's not complex at all. Look at the TRIM code in Linux, and you'll see that it's quite simple. The drivers already have routines for converting the virtual LBA to the drives' actual blocks. For RAID 1, they don't even need to do that, but pass the block on verbatim.

The problem is that Windows 7 needs an update too, before it will work with other drivers. So far, there has been no need, because there are no other (Windows) drivers that support TRIM.

 

Hi, Arth1!

To be absolutely clear, let me first start off by saying how very pleased I am with your thoughtful and detailed response - I genuinely thank you for putting such time into this! Now, let's examine additional detail. For background context, I will mention in passing that I built the first high-volume portable "PC" (which I'm aware of, anyway) that used only an "SSD" (including specifying the custom ASIC that included the base NAND Flash controller) - the OLPC machine.

Regarding RAID1 speeds, let's look at the Vaio Z specifically, since that's the context of this thread. Let's start with the benchmarks which we've all seen. The numbers would seem to show performance that indeed looks to be on the close order of 4X that of the standard Samsung MLC Flash controller. This strongly suggests that the Flash modules either support two controllers/board, or a next generation controller which seemingly adds near-zero overhead to two queued/simultaneous requests - this supports your cause.

However, the question I would ask is simple: do the present drivers actually perform in this manner? I've not yet seen any driver that is smart enough to do so - instead, current drivers (which I know of) instead issue parallel requests to write data to the "master" and "slave" drives simultaneously, or to read from the primary drive exclusively. I'd emphasize here that the primary cause of disk "failures" in RAID1 arrays is drive firmware, which wouldn't necessarily help a true RAID1 setup. Am I just out of touch with the current state of drivers (please remember that primary Flash was pretty rare back when I went there, and rotational latencies would most likely not have made this distinction a worthwhile optimization for the mainline distrib)?


Next, MTBF. I freely admit, as I tried to emphasize in my first post, selection of desired RAID type is a personal conclusion, not a be all and end all absolute. I believe based on personal experience with RAID0 (my current desktop has a 5-drive boot partition), RAID5 (the remaining partitions on that machine are configured as such), and RAID6 (my home file server is a 12-disk array) that RAID0 is entirely sufficient for SSD setups. All these are driven by Areca "H/W" controllers (where even here, the crucial S/W is just running on the board, instead of on the primary system CPU). On this point, we simply disagree.


As to kernel- versus driver-level RAID control, we're just miscommunicating. There is absolutely a correct view that states this is software, no disagreement. My point there was to distinguish between what the O.S. knows about "globally" versus that which drivers are responsible for. I've not made the distinction that you're making regarding who's written the bits. I was trying primarily to emphasize that "correct" driver performance can easily correct this (even if is Microsoft who must ship the bits).


I accept your point on the Linux view of the SSD array, since I've never run an Intel SSD driver on my Linux systems. However, the percentage of those of us who run Linux at all - especially on leading-edge machines such as the Z - is quite small. I've actually run Acronix restores on Windows RAID systems successfully. Did I just get lucky? I'll retry this soon - definitely on an Intel-based RAID array - to verify whether this can properly be repeated.


Regarding the status of H/W vendor supplied "mini-drivers" that support TRIM, we certainly agree. Until vendors get their act together, there's insufficient force (in this case, to get Microsoft) to do the right thing. From my perspective, it seemed to me that some configs like, oh, 4-drive RAID5 might be challenging to match the O.S.'s block wrap points. Perhaps I'm wrong there. If so, please accept my apologies!

Thanks again for a very enjoyable interaction!

Cheers!
Mark

 

Hi, Folks,

Just as arth1 said, there are indeed problems restoring onto Intel's RAID arrays, stock out of the box. I've now got the Acronis Plus Pack, which lets you build a custom WinPE image where you can add the Windows drivers that are appropriate for your system for the restore (this includes the full Acronis Recovery package that runs on this WinPE O.S.). They also have a neat little tool for restoring backups from Windows that'll let you dynamically patch full-disk restores onto different hardware, so that you can migrate from one type of RAID onto another type of hardware (called Universal Restore).

It does take some time, and the Plus Pack, to be able to build these custom bootable restore disks.

FYI!
Mark

 

That's encouraging news, thanks. I take it you also tried the recovery disk on your system? Is it the exact same controller as in the Z? Same hardware? SW? (There's a screenshot of the Z's controlled BIOS setup screen in the Z owner's thread.)

 

Zoinks posted an interesting screenshot of the two discs running in JBOD mode and seems to support TRIM.

http://forum.notebookreview.com/showpost.php?p=6007278&postcount=15

 

This indicates that the drives do support TRIM, but it does not mean the drives are receiving TRIM commands from the OS. The driver for the controller (in this case Intel RST V9.5.4.1001) does not currently send TRIM commands to attached drives, whether or not the drives are in RAID or JBOD.

 

Are we absolutely sure about the RAID controller...? Has it not been hacked by Sony to include TRIM..? Or something else..?

 

what about replacing the HDD with a SSD? will the raid-controller still mess up things and disable TRIM even if there is no raid or JBOD, just a single SSD disk.

 

Howdy folks,

I'm working on a test plan to see what impact long-term usage of the SSD drives do on the new Sony Z. The device I have has 2x 64GB Samsung SSD's.

I'd like anyone's feedback on the best way to artificially cause SSD performance degradation.

I'm going to fill the drive to max capacity a few times. One with large files, once with thousands of smaller files.

Any other ideas? Fill it with IOMeter? Defrag? Throw it on the ground and yell at it?

I plan to do this with drives in RAID0, then break them apart and do this to the drives individually.

 

hey ZoinksS2k. what about the anand article's method? he uses iometer, but somehow guarantees each pg is filled.

http://www.anandtech.com/storage/showdoc.aspx?i=3531&p=12

not sure if i'm understanding this myself, but his article is what we're all reading regarding degradation metrics.

 

Yeah, I think IOMeter requires an installation. This would be easier with normal drives. I could just hook it up to my workstation. The Sony SSD has a proprietary SATA connector so I have to leave it in the laptop.

If I'm going to fill the drives, I'll need a drive to boot the laptop to. Maybe I can do a XP install to USB or something.

Guess this is another example of why I need to get my Linux skills updated. I'm a windows-only guy currently.

 

usb bootdrive might do the trick. the benefit of matching his method, of course is the ability to compare to all the other drive's results for the random 4k writes -- the part that goes south without trim/gc.

hey thanks a lot for all this work. i'm about 4days from getting my z in the mail and a lot of my worries are already gone, thx to you & dimension6 & arth1

 

I'll give it a try.

I'm debating now whether or not I want to try the secure erase now and "season" the drive or hold off. I've been doing a ton of work on the drives as-is, might make sense to wait. I can always refill/season after.

 

I would say make the SSD work hard .. secure erase later ..

 

*nod*

I'm rebuilding from the recovery DVD's now.

In the "interesting" category, the recovery environment has a erase/wipe option. It lets you choose whether to write all 0's or do a few DoD-grade erasure passes. I selected the "0's" option and killed it once I saw the HDD activity light go solid. I then booted to the Windows 7 install CD and saw that it had removed all partitions/volumes from the disk (like the 'clean' command does in diskpart).

*If* a destructive rebuild is required to restore performance, this may make the process a bit easier for the average user.

 

I don't think that is anything new. The discovery discs for the old VAIO Z contains HDD erase/wipe utilities as well. However, I don't know if the erase/wipe utilizes for these new VAIO Z notebooks are "optimized" for RAID0 setups/SSDs or not.

 

New to me counts for something, doesn't it?

I've got a few versions of HDDErase to try, but my confidence isn't high. For the same reason TRIM commands may not be reaching the drive, the secure erase ATA command might not make it either.

 

You are right, from what I read HDDErase also requires the controller to be running in AHCI mode to work.

Also, the recovery option to write all 0s to the drive probably makes things worse. MLC flash requires all 1's to be written to a block for that block to be cleared. Writing all 0s actually fills up the block.

EDIT: Whoops looks like HDDErase needs IDE mode, but it looks like you already knew that.

 

Hi, Folks!

First of all, sorry for the delay! I've been migrating my most frequently used desktop from a 5-drive RAIDx configuration to a Crucial C300 256GB SATA-3 SSD drive plus Intel RAID0 HDD setup. I'll try to include the relevant caveats below, so as not to over-generalize.

McMagnus: No, to be quite clear, this is not identical hardware. However, my tests to verify arth1's assertion that Acronis has problems with restoring RAID0 drives on Intel controllers were effectively equivalent to the Z (and again, he was right on that point). Both systems use Intel's RAID controller chip, and we were talking about a process which is essentially neutral to drive type, or the specific motherboard which that controller sits on. Beyond that, as I've said, the process of creating custom WinPE disks with Acronis is quite time-consuming, to the point where I've only yet created a test CD and Flash drive with standard HDD drivers. I will say that the process is involved enough that even most enthusiasts might find it rather daunting.

Gang: The good news is that I've learned something important during this process, as long as you're a Windows 7 user that is concentrating on Windows 7 backups and restores (i.e. most relevant to this thread). Simply enough, that's that Windows 7's backup and restore facilities are surprisingly good! Not only are they quite fast, they also take care of lots of work "automagically" for you. If you do an "image backup" of the old drive(s) using the internal Backup program to a local (like USB-attached) drive, you can easily restore that backup to another drive(s). Using a standard Windows 7 install DVD, I used the Repair option to restore my previous (~~25GB) 5-disk HDD RAID0 boot array backup to the C300 SSD drive in well under 10 minutes - sorry I didn't time this. Interestingly enough, this was to a brand-new "blank" drive which I'd never previously powered on (essentially the same as having the drive(s) just HDDERASEd). After doing so, it not only just booted up, but Windows 7 had created partitions that were disk-aligned to yield "maximum" performance.

Another caveat: the C300 fully supports TRIM. As a result, I've not yet needed to run HDDERASE, so I can't tell you if I'll be able to successfully do so without unplugging the SATA cable during power on (which some motherboards seem to require, until just before running HDDERASE). However, I've learned a cool trick from the Acronis custom boot disk creation process, and I'll be trying to use a variation of this process to simplify restores for Z users who are using Intel Matrix RAID setups. I'll post if this experiment is successful.

Cheers!
Mark

 

Are you guys saying that we cannot run HDDErase on the Z's Samsung SSD's because of BIOS restrictions?

 

Hi, Sshe11,

By no means am I making such a claim, particularly when I don't have a Z in my hands to test. However, a bit of Googling will uncover the fact that, at least on some desktop motherboards, HDDERASE will only run successfully if you:

• Turn the system off
• Disconnect the SATA data cable
• Turn power on
• Boot a DOS disk which contains the HDDERASE utility
• Insert the SATA data cable while the system is running
• Run HDDERASE

Does this mean that it couldn't run here? No, but if this same restriction does apply to the Z, it could make it sufficiently difficult that few folks would be able to use it - at least on a regular basis.

We just don't know the answer, but I point this out solely because it's a frequently noted limitation.

Regards,
Mark

 

He is asking that because according to screen shots of all BIOS options for new Z there is no option to switch from ACHI to IDE an there fore it is impossible to run HDDErase. This is big problem because if for some reason SSDs become saturated, it will be hard to find the way to wipe them out.

Seams like we need small EMP generator for it

 

Hi, Igorstef!

This is purely conjecture, but I wonder if the issue I posted is essentially just a work-around for AHCI mode being set in the BIOS? Not that it would help much, but that would at least be consistent. Occam's razor, and all that. Thank you for the info!

Cheers!
Mark

 

1) Windows 7 supports trim
2) The SSDs used in the Z support trim in firmware
3) Intel RAID drivers currently do not support trim
4) Reports are that Intel is actually working to add trim support

As soon as #4 happens, this will all be moot.

 

Unless the RAID mode Intel firmware needs an update for this to happen, i.e. providing AHCI command set support in RAID mode.

 

Maybe, but as you said you self not all of us will be willing to open new Z. You can maybe use Linux to try to do secure erase using hdparm utility, but even then it is possible that BIOS will block ATA secure erase command.

Hope somebody will find solution.

 

Hi, Igorstef!

If indeed disconnecting the SATA cable is a work-around to get into IDE mode so that HDDERASE can properly identify SSD drives, then at least those who are willing to disassemble their system might be able to benefit from it - almost by definition, this weird little process would ensure that the BIOS doesn't get a chance to do much to the drive. As we agree, though, this absolutely would still limit this to a pretty small audience.

Cheers!
Mark

 

Also difficult. The connector on the SSD is proprietary. I'm going to try pulling it out of the chassis after lunch

(preventative, "That's what she said")

 

To end all this speculation, someone should try and run HDDErase on the new Z.

 

Nice. Very nice.

 

I'm almost to this point in my testing. Just a few more JBOD things to complete.

If you think the how-to I posted this morning was detailed, just wait for the performance testing.

I may struggle to get it done today, however.

 

Hi, ZoinksS2K!

You rock! Thank you very much for all the detailed work, plus spending the effort to help us all to understand what you've encountered!

Best Wishes!
Mark

 

You can help

Check here: http://forum.notebookreview.com/showthread.php?p=6016573#post6016573

gratitude and rep points to all who help