SSDs - State of the Product?

[Clocker]

I wouldn't know how to check. If you have a test I can run let me know.

I don't think TRIM is enabled on these drives. I thought the drives incorporated some type of RAID controller to acheive their high speeds and that you could not implement TRIM on a RAID-0 array.

Been a while though..

Edit: Looks like I was wrong depending on what vintage of drive you have:

http://www.anandtech.com/show/4382/computex-2011-oczs-revodrive-3-revodrive-3-x2-now-with-trim

 

[LunarMist]

Usually. Depends on the user.

Well, it's for me. I was thinking that beyond 240Gb there would not be any increase in performance or it would be negligible. I don't need more than 80GB of storage space, even with Win 7.

 

[ddrueding]

I'm now having 40GB SSDs fill just due to Win7 and Office2010 updates, even with all data kept elsewhere. 80 would be my minimum recommendation, with 160 giving you a nice quick place to work with images.

 

[Chewy509]

I'm now having 40GB SSDs fill just due to Win7 and Office2010 updates.

WTF? I knew Windows is bloated, but seriously... On my netbook with Arch Linux + GNOME 3.2 + LibreOffice + TexLive + Eclipse + Netbeans + everything else takes just over 5GB? (Heck, even on Sol11 desktop is shy over 10GB with everything installed).

While I know it's hard to get anything smaller than 120GB in a spinning drive, all entry level Ultrabooks are coming with 64GB SSDs... Gotta make it tight once you have everything installed, not to mention any data you might need.

 

[BingBangBop]

I'm now having 40GB SSDs fill just due to Win7 and Office2010 updates

I thought the only real useful purpose of the 40GB SSD's was for caching.

 

[Bozo]

I thought the only real useful purpose of the 40GB SSD's was for caching.

Or three of them in RAID 0

 

[Mercutio]

I thought the only real useful purpose of the 40GB SSD's was for caching.

40GB is gobs of space for an OpenSUSE installation.

 

[LunarMist]

I would not trust a Samsung drive with my apps though.

 

[ddrueding]

Yup, I was running 3x16GB when SSDs first came out. Now I let the RevoDrive handle the RAID goodness.

 

[Handruin]

Looks like Super Talent may be entering the market for the PCIe SSD breed of drives. I'll be curious to see what each of the sizes costs. My guess is slightly below OCZs RevoDrives to try and take some of their market share...unless if they end up with a real performer and blow them away.

 

[CougTek]

In the same line of ideas, Micron launched a new series of PCIe SSD for enterprises. Both freakingly fast and freakingly expensive. The speed is unheard of, at least from what I read so far. The price is quite high : ~2800$ for the 175GB model, ~5600$ for the 350GB one.

 

[CougTek]

BTW, is it possible to make a stripe of PCIe SSDs? I ask in case Santilli plans to buy some soon.

 

[ddrueding]

BTW, is it possible to make a stripe of PCIe SSDs? I ask in case Santilli plans to buy some soon.

It is. I've done it with the Revo and Velo drives. It was quicker, but not by as much as I was hoping for. The benches should be around here somewhere.

 

[Santilli]

If I had it to do over, I'd probably go with a single X-25M 120 gig as the boot drive, rather then raiding the two of them. That said, no problems so far, other then having to rebuild once. I'm pretty sure that at 250 MB/sec, with absurd access times,
other components are limiting speed, not hard drives.

 

[Handruin]

If I had it to do over, I'd probably go with a single X-25M 120 gig as the boot drive, rather then raiding the two of them. That said, no problems so far, other then having to rebuild once. I'm pretty sure that at 250 MB/sec, with absurd access times,
other components are limiting speed, not hard drives.

Sadly no...your SSD hard drive is still likely the slowest component for an IO device. The speed of your RAM would still make it look stupid.

 

[ddrueding]

It is still likely the slowest part of your IO stream, but the percentage of time that your system is waiting on the drive has shrunk massively. The gain from further improvements is nearing zero as soon as you go SSD in the first place (for normal usage).

 

[Santilli]

"The gain from further improvements is nearing zero as soon as you go SSD in the first place (for normal usage). "

Exactly.

I haven't been able to notice any difference going from a raid 0 three drive array to a single drive, using the Vertex Turbos.

I suspect David has it down going to the OCZ RevoDrives to notice any real improvement, or noticeable improvement.

As for the IO of this machine:

IIRC I created a ram drive, and tested it. It was running around 3500 MB/sec, IIRC.

What are the Revos running at? 800 MB/sec?

 

[ddrueding]

Give or take, depending on which one you are running. They'll do 1600MB/s on the high end.

 

[Santilli]

Besides booting, do you notice any difference vs. the two X-25 M's?

 

[ddrueding]

Sure. Pano stitching is awesome, generating thumbnails on RAW files is great, and level loading in games positively rocks. But I've also made considerable changes in CPU since the X-25 days.

 

[Handruin]

It is still likely the slowest part of your IO stream, but the percentage of time that your system is waiting on the drive has shrunk massively. The gain from further improvements is nearing zero as soon as you go SSD in the first place (for normal usage).

I have no doubt an SSD is fantastically faster than a traditional HDD. I was addressing Santilli's statement that he said he was pretty sure other components were holding back the system, not the SSD. I don't believe that to be true.

Also, 250MB/s or 800 MB/s, etc is far too generic of a claim to really mean anything. Was that with a 4KB random write or a 8MB sustained transfer read? My point is, you can't judge the entire performance of your disk IO subsystem from that one single measurement.

 

[Handruin]

"The gain from further improvements is nearing zero as soon as you go SSD in the first place (for normal usage). "

Exactly.

I haven't been able to notice any difference going from a raid 0 three drive array to a single drive, using the Vertex Turbos.

I suspect David has it down going to the OCZ RevoDrives to notice any real improvement, or noticeable improvement.

Exactly the reason why you can't judge the entire disk subsystem from a single sustained transfer rate benchmark value.

The revo drive also removes (if I'm not mistaken) the SATA II bus as a limiting factor by giving it access to the PCIe bus and what is the difference in IOps between it and your SSD?

Just because you anecdotally dont notice a difference doesn't also mean you're not stressing your PC with your basic tasks.

 

[Clocker]

I think the problem is that most people are not doing anything disk intensive enough to make the increase in speed from a single SSD to a RAID setup very noticeable. Santilli has reached the point of diminishing returns with his computer workload. It's hard to notice differences of a 1/10 of a sec. Basically it comes down to what you're doing with your PC or the benchmark you're running. The disk subsystem is still the bottleneck, it's just plenty fast enough for what most people are doing.

 

[time]

If an SSD can service 50,000 IO requests per second, that's 20 microseconds per request.

A fast SSD can sustain (read) transfers of 500 MB per second for very large requests (eg 8 MiB), which works out to 16,000 microseconds each or 8 microseconds to transfer a single 4 kiB block.

So the total service time for a 4 kiB request is 28 microseconds, or about 147 MB per second (in the ideal world).

Ignoring overheads, striping 2 drives in RAID 0 yields (20 + 7.8/2) = 23.9 microseconds or 171 MB per second. That's only 14% less time spent waiting for disk I/O.

This is all wildly optimistic because writes are an awful lot slower. Don't forget that the OS needs to regularly flush data cache writes to avoid the risk of massive data corruption.

The bottom line is that about the only time I'd expect to see a significant benefit from desktop SSD RAID-0 is when copying files.

I know this is a gross simplification - anyone care to refine it?

 

[LunarMist]

For most purposes a single fast SSD is better than several mediocre SSDs in RAID 0.

 

[BingBangBop]

I thought it would be good to document the new OCZ RevoDrive 3 here by including a review.

 

[LunarMist]

Meh. It needs trim. I like the Cherrydale and the Intel software that can optimize the drive.

 

[Bozo]

The latest Intel software I found still won't Trim a RAID setup. Am I missing something?

 

[LunarMist]

No, I don't think so.

 

[LunarMist]

Where is that promised Intel PCIe SSD? It has been forever.

 

[ddrueding]

Where is that promised Intel PCIe SSD? It has been forever.

Indeed. Anyone finds one let me know!

 

[LunarMist]

Until then I'll live with the two Cherrydales and the Inferno. RAID 0 is not doing much according to the stopwatch.

 

[BingBangBop]

Intel introduces 910 Series PCIe SSD

An enterprise level SSD 400-800GB in size using an 8x PCI-e slot getting up to 2GB/s using sequential reads or 180,000 IOPs for between $1,929. and $3,859. MSRP.

 

[ddrueding]

Sweet. And at under $5/GB...not bad in that performance segment. I have yet to put a storage subsystem that expensive in a workstation, and I doubt ESXi supports it.

 

[Mercutio]

Micron just released a new mini-PCIe SSD. SR has a review posted. I'm glad that form factor is still being supported, particularly since it looks like a considerable improvement over my current Intel 310.

 

[ddrueding]

Grr. I just ordered the 120GB OCZ because it was the biggest mSATA on Newegg...

 

[Mercutio]

Apparently it will be a while before the Micron drive will be available to consumers anyway.

 

[LunarMist]

Intel introduces 910 Series PCIe SSD

An enterprise level SSD 400-800GB in size using an 8x PCI-e slot getting up to 2GB/s using sequential reads or 180,000 IOPs for between $1,929. and $3,859. MSRP.

Damn. I've been waiting a long time for that, but $4K is a bit much.

 

[Chewy509]

Damn. I've been waiting a long time for that, but $4K is a bit much.

Anandtech has a review on it as well, but they mention the Intel 910 is NOT bootable.

The use of LSI's 2008 SAS PCIe controller makes sense as there's widespread OS support for the controller, in many cases you won't need to even supply a 3rd party driver. The 910 isn't bootable, but I don't believe that's much of an issue as you're more likely to deploy a server with a small boot drive anyway. There's also no support for hardware encryption, a more unfortunate omission.

Source: http://www.anandtech.com/show/5743/intels-ssd-910-400800gb-mlchet-pcie-shipping-in-1h-2012

 

[ddrueding]

Not much of an issue? That sucks.