TRIM in RAID-0 on ICH9R? Any hope?

[VirtualLarry]

Is there a driver version for ICH9R RAID-0, that works and send TRIM to the members of the RAID-0 array? (Potentially four 50GB Vertex 2 drives.)

OS is Win7 64-bit

 

[aigomorla]

no u wont ever see trim working on R0
no dont need it either... speaking from experience on using R0 SSD's for quite some time now... Gen1 Intel...

to be honest, your will most likely upgrade to a larger cap ssd b4 u even hit 60% life even on RAID-0 unless u run a private data center.

OCZ might have its own software, however im not too sure.

Garbage collection works outside of trim and Raid... its built native onto the SSD.

 

[Sunburn74]

I thought a work around to enable trim for raid 0 was published here on anandtech about 6 months ago...

 

[Old Hippie]

I thought a work around to enable trim for raid 0 was published here on anandtech about 6 months ago...

Not for that chipset.

There will be no TRIM for the ICH9 or ICH10 chipsets.

 

[sub.mesa]

no u wont ever see trim working on R0

Why not? ;-)

no dont need it either... speaking from experience on using R0 SSD's for quite some time now... Gen1 Intel...

'Need' is relative, but you would want TRIM support. Without it, both write amplification and performance get worse over time. Garbage collection is no replacement for TRIM; only overprovisioning can be a replacement for TRIM. Usually consumer drivers have very little overprovisioning built into the difference between GiB and GB.

RAID and TRIM are not enemies as well; Linux and BSD are doing it for years. Even ZFS support TRIM, which is a lot more complicated.

On the Windows platform, the old architecture is showing and iRST supported TRIM on Windows 7 but not on Windows 8. And only with specific chipsets, starting with Z68. However, this is just a political decision from Intel. They could have just as easily enabled it for other chipsets. I also remember that some people hacked the drivers with probably a hex editor or something, and managed to get it working on older chipsets like the 50-series.

 

[aigomorla]

Why not? ;-)
'Need' is relative, but you would want TRIM support. Without it, both write amplification and performance get worse over time. Garbage collection is no replacement for TRIM; only overprovisioning can be a replacement for TRIM. Usually consumer drivers have very little overprovisioning built into the difference between GiB and GB.

RAID and TRIM are not enemies as well; Linux and BSD are doing it for years. Even ZFS support TRIM, which is a lot more complicated.

On the Windows platform, the old architecture is showing and iRST supported TRIM on Windows 7 but not on Windows 8. And only with specific chipsets, starting with Z68. However, this is just a political decision from Intel. They could have just as easily enabled it for other chipsets. I also remember that some people hacked the drivers with probably a hex editor or something, and managed to get it working on older chipsets like the 50-series.

because TRIM needs to be passed though the ICH.
For TRIM to be passed though the ICH, it requires a newer ICH.
The fact the OP is using such an old ICH.... not even ICH10R, let alone the PCH your talking about in SB, IVY, and HASWELL, you want to tell the OP its possible?




And i said TRIM wont be missed on a R0 system with SSD's.
You would need to do a massive amount of writes per day to impact the life on a SSD.

Chances are you R0'd for space on the SSD more so then speed, because if it was for pure speed, you failed. R0 on SSD's show almost 0 visual improvement.

If you R0'd for space, well, i pretty much gaurentee you will get greedy with more space b4 ur SSD's in R0 see even 60% life even without TRIM.

 

[Old Hippie]

the old architecture is showing and iRST supported TRIM on Windows 7 but not on Windows 8

But not in any RAID mode.

I have 3 256GB C300s attached to a LSI 9460-4i card and my performance has stayed the same.

This is from as-ssd-bench Volume0 12.7.2011 6-38-45 PM

[/URL][/IMG]

This is from as-ssd-bench LSI MR9260-4i SC 11.12.2013 1-02-19 PM....

[/URL][/IMG]

 

[sub.mesa]

because TRIM needs to be passed though the ICH.
For TRIM to be passed though the ICH, it requires a newer ICH.
The fact the OP is using such an old ICH.... not even ICH10R, let alone the PCH your talking about in SB, IVY, and HASWELL, you want to tell the OP its possible?

Absolutely. It is a software issue not a hardware issue. You can pass TRIM through an old SATA/150 just fine, even PATA would not be a problem if the controller just passes commands it receives. In other words, there doesn't need to be any hardware support. It does require proper support from the software, which sends the ATA commands, as well as the device which receives it.

For example, if you boot into Linux or BSD you can use TRIM just fine on older systems. Mac OSX also supports TRIM even on the older Intel chipsets, though it only enables it for Apple-shipped SSDs. This can be tweaked so it works with other vendor SSDs as well.

Windows is another story. It has an old storage backend, meaning it cannot properly support TRIM and UNMAP (the SCSI equivalent to TRIM). There is UNMAP support in Windows 8 however. I have not tested it, but the problem is that TRIM/UNMAP are not unified, as it should be. There will still be many cases where you will not have TRIM support.

And i said TRIM wont be missed on a R0 system with SSD's.
You would need to do a massive amount of writes per day to impact the life on a SSD.

The amount of writes does not directly affect the need for TRIM. If you only do large writes you would never need TRIM. If you do random writes alot, then overprovisioning is pretty much mandatory to prevent the write amplification from rising; shortening the lifetime of the SSDs needlessly and can also impact performance. The latter is less prevalent on SSDs with aggressive garbage collection, such as Samsung. But it should be noted that this will further increase the write amplification, which might not be a good idea if you got a TLC-based Samsung SSD.

R0 on SSD's show almost 0 visual improvement.

Define visual improvement please. Do you mean improvement that can be subjectively perceived by the user during real workloads, or do you mean that it can be perceived through testing/measuring/benchmarking? The latter is very much true.

You should also know that SSDs are one giant RAID0 already; a single SSD can be described as 16 tiny SSDs in RAID0. Or actually RAID5 since modern SSDs utilise parity correction in addition to ECC error correction.

I do agree with you, however, that with typical light desktop usage there would not be an unacceptable penalty to lifetime as well as performance. But it might in other cases. But example, people who buy a 64GB SSD for SRT caching and use up the full 64GB without leaving any space for overprovisioning. They should have bought 120GB of course. Due to the excessive number of random writes and lack of space space, the SSD will wear faily quick and might actually fail within a reasonable time (<3 years).

I always advise to put a little bit of overprovisioning. You can always undo it in the future by extending the partition to a larger size, which can be done by Windows' native Disk Management. So start with a lot of OP and shave off what space you need might be a good strategy for those who want to play it safe.

 

[Cerb]

You should also know that SSDs are one giant RAID0 already; a single SSD can be described as 16 tiny SSDs in RAID0. Or actually RAID5 since modern SSDs utilise parity correction in addition to ECC error correction.

Only a subset of Sandforce drives (those not advertising 2^nGB sizes), and the Crucial M500, do this, AFAIK, when it comes to consumer SSDs.

 

[sub.mesa]

Intel as well since the 320 series (and of course newer S3500/S3700 series). However, Intel uses a dedicated NAND die for parity, so it behaves more like a RAID4 as opposed to distributed parity in RAID5. This doesn't change much however, the working is similar.

Samsung appears to be an exception. In addition to missing parity correction, they also lack capacitors to protect against sudden power loss. In a sense they're still at the level where Crucial M4 was, in terms of data protection.

 

[Fernando 1]

Is there a driver version for ICH9R RAID-0, that works and send TRIM to the members of the RAID-0 array?

What you will need is a modded Intel RAID ROM version, which will let TRIM pass through the Intel ICH9R SATA RAID Controller. Last year Dufus has developed an especially modified Intel OROM v11.6.0.1702 with TRIM in RAID0 support for Intel's 4-Series Chipsets. Meanwhile this modded Intel OROM has been successfully tested with Intel ICH10R RAID0 systems (look >here<). So it may be possible to get the TRIM in RAID0 feature even with an Intel ICH9R system.
Additionally you will have to run an actual Intel RAID driver of the v11 or v12 series.

 

[sub.mesa]

Sounds plausible, since the Option ROM modification is probably to trick the drivers into believing it may utilise TRIM. Very curious if it actually works on your old ICH9. Somehow I think that the new (hacked) Option ROM might not work on ICH9 at all.

 

[Cerb]

Samsung appears to be an exception.

It's the other way around. For consumer drives, they're either Sandforce without RAISE disabled, Crucial M500s, or they don't have that feature. Samsung, Toshiba*, Sandisk*, Plextor/Lite-On, and others, don't.

Parity protection is not standard, and is not likely to become standard, at least not for consumer and SMB SSDs.

* They still offer some models with Sandforce controllers that have RAISE enabled, but also do offer, or have offered, Sandforce models with RAISE disabled. Current-gen models don't have any such thing going on, however.

 

[sub.mesa]

Not sure how many Sandforce-powered SSDs lack RAISE. I would think most of them do support it? In most cases, I can see it is disabled on the 60/64GB or 120/128GB version while being enabled on the higher capacity models.

There is a lot of junk on the market, but out of all the popular models advised on forums like these, I would think Samsung is the only big exception to the generale rule that parity protection is getting standard, except for the cheap/lowend SSDs.

 

[Cerb]

Not sure how many Sandforce-powered SSDs lack RAISE.

Any with an 2^nGB advertised price lack it: 64GB, 128GB, 256GB, 512GB.

There is a lot of junk on the market, but out of all the popular models advised on forums like these, I would think Samsung is the only big exception to the generale rule that parity protection is getting standard, except for the cheap/lowend SSDs.

Except that's not the case at all.
Sandisk Ultra Plus: none.
Sandisk Extreme II: none.
Toshiba Q drive (THNSNH): none.
Seagate 600: none.
Any Plextor: none.
Most OCZs: none.
...

It's not a feature that's going away, but it's not a standard feature, general rule, etc.. Crucial and Intel (on the their S3500 and S3700), FI, make a point about having it in their literature, because it differentiates them for business customers.

 

[Old Hippie]

What you will need is a modded Intel RAID ROM version, which will let TRIM pass through the Intel ICH9R SATA RAID Controller. Last year Dufus has developed an especially modified Intel OROM v11.6.0.1702 with TRIM in RAID0 support for Intel's 4-Series Chipsets. Meanwhile this modded Intel OROM has been successfully tested with Intel ICH10R RAID0 systems (look >here<). So it may be possible to get the TRIM in RAID0 feature even with an Intel ICH9R system.
Additionally you will have to run an actual Intel RAID driver of the v11 or v12 series.

Cool!

Ya learn something new everyday!