SSD in RAID

[Will Rickards]

Intel's new 320 series seems to be getting reviewed today.
Not sure of the availability.
Are they in your price/performance range?
Link

 

[Bozo]

Apparently, another Sandy Bridge problem?

 

[ddrueding]

Possibly. My motherboard is only SATA II, 3GB/s.
Your REVO is plugged into a PCI-e slot, correct?

Yup. Dedicated 4x PCIe.

 

[time]

This is the thread that details the issues. Reports of stuttering, halved speed. Possibly dependent on which controller you use.

After running some searches on that forum, the only clear information I could find that is that stuttering afflicts people with Marvell motherboard controllers - for them, the problem goes away if they use the Microsoft AHCI driver. Even that observation may be out of date.

 

[time]

I'm only a couple of days away from pulling the trigger on the first (perhaps the only ) of these configurations.

So it seems that Intel 320 (G3) turned out to be G2.1, and what planet are Intel currently on?

I still like the idea of Crucial/Micron and buying direct. Only slightly uneasy about the firmware; AFAIK it's developed inhouse which is a significant improvement over the Sandforce rebadgers. We're now up to Crucial/Micron M4/C400 rather than C300, which has me torn. For non-RAID, I'm leaning towards the newer M4, even though my conservative side is screaming at me that it's unproven. I'm really vacillating on the RAID side; on the one hand the C300 is relatively proven (v6/v7 firmware), but on the other it's always best to have the newest when it comes time to replace failed units.

On top of all that, I just read a thread about anecdotal stories of premature SSD failures. It included all brands; OCZ was more prominent but that's probably because they sold a sh*tload of them. Intel seemed to be crap for 80GB but angelic for 160GB ...

So I guess I have no idea.

 

[ddrueding]

It is all anecdotal. I've bought tons of SSDs, and the only failure was clearly abuse.

 

[Bozo]

SSD's in a RAID configeration don't run TRIM. I'll stay with single drives for now.
Another thing: I really don't think SSDs are ready for prime time. Why are all the TRIM utilities different? Aren't there any specifications or standards that everone is supposed to follow? On a spinning disk, defrag is defrag and check disk is check disk no matter what hard drive is installed. Performance tuning might be proprietary but basic utilities should all be the same. The crucial forum has post that have owners complaining that their SSDs won't work correctly with the controller chip on their motherboard. What's up with that? No standards?

 

[ddrueding]

Checkdisk still works and defrag is unnecessary. Just get a good SSD (with built-in garbage collection) and forget about trim all together.

It isn't complicated.

 

[time]

I don't believe TRIM is necessary unless you're writing serious amounts of data, eg. benchmarking. On most SSDs, garbage collection is designed to deal with normal loads (aggressive GC increases 'write amplification' which may impact endurance, although again, probably not an issue in most real situations).

I've seen lifetime endurance figures of 30TB of writes bandied around for the latest generation of MLC drives. Evenly distributed over 5 years, that's 685MB per hour or 50 4kB blocks per second, continuously. Offhand, I think the popular AS SSD benchmark writes more than 5 gigabytes in a couple of minutes.

 

[time]

Why are all the TRIM utilities different? Aren't there any specifications or standards that everone is supposed to follow?

TRIM is a standard ATA command, so it's the same for all drives. But Windows reserves hardware control for itself, so circumventing that (in Windows XP) can be a challenge. In Windows 7, the utilities are simply creating a large file and deleting it, which prompts Windows to issue a TRIM command.

The crucial forum has post that have owners complaining that their SSDs won't work correctly with the controller chip on their motherboard. What's up with that? No standards?

I've trawled those forums and didn't find anything that really worried me - unless I was an Apple owner, in which case I'm screwed by Apple's consistently poor adherence to industry standards. Do you have any examples? I may have been looking at it through rose colored glasses.

 

[Mercutio]

I've seen lifetime endurance figures of 30TB of writes bandied around for the latest generation of MLC drives.

It's been suggested to me very strongly that any reliability figures for MLC drive coming from drive OEMs have only casual relationship with reality. I've mentioned that in the other SSD thread but perhaps it might be wise to consider a drastic downward revision of that 30TB number just on the basis of our own lack of first-hand experience at wearing out the drives?

 

[time]

Micron rates current 25nm flash for 3,000 erase cycles. Intel rates exactly the same stuff from the same factory for 5,000 cycles.

I guess Dilbert works for Intel these days.

Anyway, that works out to erasing a block of 128 pages once for every 10 4kB page writes. Unless you're doing nothing else other than random 4kB writes, that doesn't seem completely unreasonable to me.

The stat was for a 128GB SSD; a 256GB SSD would halve that to one block erase for every 5 pages.

Does that fit with what Deep Throat is saying?

 

[Bozo]

TRIM is a standard ATA command, so it's the same for all drives. But Windows reserves hardware control for itself, so circumventing that (in Windows XP) can be a challenge. In Windows 7, the utilities are simply creating a large file and deleting it, which prompts Windows to issue a TRIM command.

If the Trim command is a standard, why doesn't Intels toolbox work on a Crucial SSD?


I've trawled those forums and didn't find anything that really worried me - unless I was an Apple owner, in which case I'm screwed by Apple's consistently poor adherence to industry standards. Do you have any examples? I may have been looking at it through rose colored glasses.

You might be right about the apples. I believe I saw a couple of mentions about some AMD motherboards too. I didn't pay a lot of attention as they were not what I was looking for.

 

[time]

If the Trim command is a standard, why doesn't Intels toolbox work on a Crucial SSD?

Because it's using some sort of backdoor to issue Intel-specific commands to the SSD. Only ATA commands are standard.

 

[sor]

I attached one SSD to the 9650se. Then ran AS SSD benchmark. The speeds were one third of what they were before I attached the SSD to the 9650se.
I need a new controller

We've got a few hundred 9650s deployed, they're decent, but beginning to show their age. I'd be curious, are you running with cache? You may be paying a latency penalty (yes I realize this was two months ago).

 

[sor]

I'm not sure that I still believe in the need for TRIM, because I suspect idle garbage collection is sufficient in most real world situations (not sure about Sandforce though). Anand hammered a C300 with writes, waited just 3 hours, then retested. Unsurprisingly, it had only recovered slightly. :roll:

Yeah, that bugs me. It's called "idle time garbage collection" for a reason. I suppose it's a good indicator for busy server use, but a desktop is going to have idle disks 90% of the time... plenty of time for these modern controllers to clean things up.