Need SSD for *very specific* usage scenario - Advice?

[VirtualLarry]

So does anyone know which drives will perform GC under non-idle conditions outside of large writes? That is the question I'm hoping someone can answer.

I heard good things about the Kingston V+100 drives being good for windows xp, because it lacked trim, and those drives had good GC.

Unfortunately, they are no longer carried new at Newegg any more.

 

[thtadthtshldntbe]

Rare poster here, but wouldn't what you want to do be better suited with some sort of storage server type solution, like something running ZFS on actual pc hardware, that would not be overwhelmed by the number of reads or disc accesses.

You could set up one storage server and have all the devices writing to it and do your reads from there. This setup would have other advantages like data redundancy and error correction. It would also make backing up the data easier. And the combined total cost of SSDs for all your machines could easily cover assembling such a server.

 

[TrackSmart]

Thank you for the range of insightful comments.

To jwilliams4200: I did not mean to dismiss your drive recommendation or solid advice. I was still hoping someone would know of a non-enterprise SSD with garbage collection algorithms that would work well under these conditions. If getting controller X over controller Y means less maintenance and instrument down-time (i.e. drive removal), that's a big win in my book and worth investigating.

It's possible that ALL modern, consumer-grade SSDs would be fine. I don't know. That's why I am here, asking the saavy readers of Anandtech. Maybe someone else has a similarly constant workload and runs consumer drives? Maybe there is a SandForce of Marvell employee hanging around here. Who knows?

 

[TrackSmart]

Rare poster here, but wouldn't what you want to do be better suited with some sort of storage server type solution, like something running ZFS on actual pc hardware, that would not be overwhelmed by the number of reads or disc accesses.

You could set up one storage server and have all the devices writing to it and do your reads from there. This setup would have other advantages like data redundancy and error correction. It would also make backing up the data easier. And the combined total cost of SSDs for all your machines could easily cover assembling such a server.

You may be right, though I'm not sure how I would implement this. We are dealing with proprietary software. They don't give out installation discs with these instruments. We clone our hard disks in case of failure, otherwise you have to wait for the company to send you a drive with the software pre-installed. Maybe I should be talking to the makers of the instruments to ask for improved software, rather than seeking hardware that works around the poorly-written software.

 

[Smoblikat]

You may be right, though I'm not sure how I would implement this. We are dealing with proprietary software. They don't give out installation discs with these instruments. We clone our hard disks in case of failure, otherwise you have to wait for the company to send you a drive with the software pre-installed. Maybe I should be talking to the makers of the instruments to ask for improved software, rather than seeking hardware that works around the poorly-written software.

What about my idea

 

[Cerb]

One thing, if you could, would be to just get one and see how it does. After some months, if it's still OK, pull it out and check what it thinks its lifespan and WA are (easiest to do with Intel drives). Depending on what it looks like then, replace them all with something similar (or not).

It's possible that ALL modern, consumer-grade SSDs would be fine. I don't know. That's why I am here, asking the saavy readers of Anandtech. Maybe someone else has a similarly constant workload and runs consumer drives? Maybe there is a SandForce of Marvell employee hanging around here. Who knows?

Not all, for sure, but maybe most.

A Marvell or Sandforce person couldn't help you, either. While drives with certain controllers have certain behaviors, there's a wild difference even between what should be nearly identical Sandforce drives, with canned Sandforce firmware. Marvell's controllers are basically CPUs with FTLs, and a prototype style firmware to start working from, AFAIK. By the time it is ready for users, it is truly Micron's, Corsair's, Plextor's, etc..

With time and Moore's Law, we'll get there. But, today, the technology is still close to bleeding edge. Keeping performance and longevity high (something like that discontinued Kingston would surely have sky-high WA, FI), with pipsqueak processors, is not gonna be easy.

 

[TrackSmart]

To Smoblikat: Yes, that would be great, but we have limited control over the software and hardware.

To Cerb and others: Yep, it's likely that I'll just have to test one of these drives over an extended period of time and see how it does. I personally gravitate towards the Intel/Crucial/Samsung models for their better track records of reliability. Samsung seems to be the recommended choice for non-TRIM environments, so I will take the advice of others and start there, unless further information becomes available.

 

[jwilliams4200]

To Cerb and others: Yep, it's likely that I'll just have to test one of these drives over an extended period of time and see how it does.

Or, as I suggested, put it into a worst-case performance state before you install it into your system. That way you will know quickly whether the worst-case performance is acceptable.

 

[TrackSmart]

Or, as I suggested, put it into a worst-case performance state before you install it into your system. That way you will know quickly whether the worst-case performance is acceptable.

Yes, that would be a much quicker way to go about things. Thanks.

 

[Phynaz]

Academic research, scientific instruments that cost thousands if not tens of thousands of dollars, and you a $200 budget for storage?

This research isn't school sponsored is it? Are you paying for this out of your pocket?

BTW, software doesn't support trim, Operating Systems do. What OS is running on the data collection box?

 

[TrackSmart]

Academic research, scientific instruments that cost thousands if not tens of thousands of dollars, and you a $200 budget for storage?

This research isn't school sponsored is it? Are you paying for this out of your pocket?

BTW, software doesn't support trim, Operating Systems do. What OS is running on the data collection box?

I came here for friendly advice, but I seem to be getting some attitude. I'm not sure why.

I won't go into the details of our funding, but suffice it to say that we do a lot of work on a relatively small budget. Important work, in my estimation. Please realize that it is rare to get full funding on a grant in my field, you typically get 60% or 70% and have to scale back plans accordingly. Or get creative. Or both. Not every instrument runs a standard operating system. Ever use a typical scientific datalogger? Certainly, it has an operating system. In fact, each model from each company typically has a different operating system labeled with an arcane combination of letters and numbers (e.g. CR1000.std.13).

If you have useful advice, I'm happy to hear it. At the moment, it seems that the details of when garbage collection happens with modern SSD controllers are not disclosed or well-documented.

 

[Caerid]

Garbage collection on all ssd's out now are pretty relaxed. They were at one point pretty aggressive just before trim was introduced. I know some of the older ocz drives you could manually force gc with a program.

If you were using a windows as an os for your data collection I could easily recommend samsung 830's because their drives include a tool, ssd magician where you can manually perform garbarge collection.

Other than that the only solution that I can think of is to buy as big an ssd that you can, preferably 512gb versions so that it doesn't run into the diminishing returns as quickly and hopefully you'll be performing full wipes before it gets to that point.

 

[StarTech]

The fact that you clone the vendor drive, makes me wonder that it is very likely the SSD partition will not be properly aligned. That would multiply the wear rate.
You can try to clone and shift partition to properly align and see if it still boots.

Otherwise, an SSD writing a small file every few seconds will be mostly iddle. You would have a hard time measuring busy time. While the current GC may be relaxed, it is hard to believe that the GC design will wait until all blocks are dirty. At some threshold point GC will kick in and collect in between writes.

 

[StarTech]

FYI, the SAMSUNG support page has an option for email or live chat with tech support. You don't loose anything by asking about GC in a no TRIM environment with small writes every few seconds.

 

[Blain]

For about $61 bucks you can buy the 160GB WD Scorpio Black and not have to worry about garbage collection, alignment or any other SSD quirks.
$61 and it's done!

* Great reputation for reliability... Low platter density... Far cheaper price...

 

[TrackSmart]

Startech, it is possible that most modern SSDs would start performing GC duties before the drive got that dirty. Or that a couple of seconds of idle between writes would be enough time to do some maintenance. That's what I would hope for. Our only experience with SSDs in these systems are early models from a previous generation. I could certainly chance an email to 'tech support' at Samsung, Crucial, and Intel to ask for advice - they'd have to pass that message to someone with considerable knowledge - which I doubt will happen. You never know.

For about $61 bucks you can buy the 160GB WD Scorpio Black and not have to worry about garbage collection, alignment or any other SSD quirks.
$61 and it's done!

* Great reputation for reliability... Low platter density... Far cheaper price...

To Blain: We currently have ordinary hard disks in the instruments (this is what replaced the early SSDs). They 'work', mostly, but cause the issues outlined in the original post. The question is which modern SSDs would cut it. They should all have algorithms that prevent the disasters we had with early SSDs, but I don't know for certain.

Thanks for your time folks! I think it's time to send a few emails and then grab a drive and test it.