How to pick SSDs
December 27, 2011 2:12 AM Subscribe
How do I decide what the best (balancing reliability and speed) SSD (SATA II) is, for x dollars?
For myself, I'd go for the one with the biggest warranty/reliability. They are all pretty fast. So if most of them are 3 years and some of them are 5 years, the 5 year ones are automatically a better deal unless they are more than 5/3 the price of the 3 year ones. (In other words, amortize the cost over the warranty period for what it costs per year that you are guaranteed to have a working hdd.)
When I need to do analyses like this, I'll set up a spreadsheet. The different models down the left, with columns for the different metrics. For HDDs, I usually include warranty, latency, sustained throughput, $/gb. Then, if I'm feeling fancy, I'll use conditional formatting to highlight the cell of the drive that "wins" each category. Whichever one has the most wins, wins.
For SSDs, I would heavily weight toward drives with fast write times. When I experimented with using a CF card as a HDD, the write speed was the bottleneck. I was surprised at how often operating systems actually do write to the disk. It wasn't too big of a problem during normal operation, but when I had to install a program, or when it was downloading updates, the whole system bogged down. (I know a CF isn't an SSD, but the technology is the same.)
What I would also do is hedge my reliability bets and come up with an easily restorable backup system. What I did is make a disk image of the whole drive, and then do "incrementals" of my user data using a robocopy script. The robocopy script only copies newer files, so that runs petty fast after the first one. My backups are on a linux server, and what I did was use hard links on the server to maintain multiple "snapshots" of my data. When the robocopy script runs, it updates a file that the linux server uses as a flag. Each night a script runs on it to rotate the backups. The script compares the flag file in the "current" backup to the flag in the "previous" backup. If they are the same, it does nothing. If different, it discards the oldest copy of the data and moves the rest down the line (I use "current", "previous", "older" and "oldest" as my generations.)
If there is a disk failure, I can rewrite the whole disk image to the new drive, and then reverse the robocopy script to replace anything on the image that's newer.
If I was feeling particularly fancy, I'd make a script that mounted the disk image and copied over the "current" data in the background, so that when I reimage the disk, it is pretty much up to date.
posted by gjc at 5:55 AM on December 27, 2011 [1 favorite]
When I need to do analyses like this, I'll set up a spreadsheet. The different models down the left, with columns for the different metrics. For HDDs, I usually include warranty, latency, sustained throughput, $/gb. Then, if I'm feeling fancy, I'll use conditional formatting to highlight the cell of the drive that "wins" each category. Whichever one has the most wins, wins.
For SSDs, I would heavily weight toward drives with fast write times. When I experimented with using a CF card as a HDD, the write speed was the bottleneck. I was surprised at how often operating systems actually do write to the disk. It wasn't too big of a problem during normal operation, but when I had to install a program, or when it was downloading updates, the whole system bogged down. (I know a CF isn't an SSD, but the technology is the same.)
What I would also do is hedge my reliability bets and come up with an easily restorable backup system. What I did is make a disk image of the whole drive, and then do "incrementals" of my user data using a robocopy script. The robocopy script only copies newer files, so that runs petty fast after the first one. My backups are on a linux server, and what I did was use hard links on the server to maintain multiple "snapshots" of my data. When the robocopy script runs, it updates a file that the linux server uses as a flag. Each night a script runs on it to rotate the backups. The script compares the flag file in the "current" backup to the flag in the "previous" backup. If they are the same, it does nothing. If different, it discards the oldest copy of the data and moves the rest down the line (I use "current", "previous", "older" and "oldest" as my generations.)
If there is a disk failure, I can rewrite the whole disk image to the new drive, and then reverse the robocopy script to replace anything on the image that's newer.
If I was feeling particularly fancy, I'd make a script that mounted the disk image and copied over the "current" data in the background, so that when I reimage the disk, it is pretty much up to date.
posted by gjc at 5:55 AM on December 27, 2011 [1 favorite]
Response by poster: I guess my question is not framed well. Let me try again:
A lot of reviews out there read like ads and don't really have much in the way of useful quantitative data — just spec sheets from the manufacturer, which are generally useless because everything they make is fast, which makes comparisons difficult, or everything performs optimally up to certain numbers, which rarely reflect real-world usage.
Is there a resource where this data is collected by price range, or true performance, etc.? Basically, I want to say, this is my budget, and the existing "spreadsheet" will say, these are the drives to look at. Something like a Consumer Reports for SSDs?
posted by Blazecock Pileon at 6:12 AM on December 27, 2011
A lot of reviews out there read like ads and don't really have much in the way of useful quantitative data — just spec sheets from the manufacturer, which are generally useless because everything they make is fast, which makes comparisons difficult, or everything performs optimally up to certain numbers, which rarely reflect real-world usage.
Is there a resource where this data is collected by price range, or true performance, etc.? Basically, I want to say, this is my budget, and the existing "spreadsheet" will say, these are the drives to look at. Something like a Consumer Reports for SSDs?
posted by Blazecock Pileon at 6:12 AM on December 27, 2011
Honestly, anandtech has already put together most of the data you want: http://anandtech.com/bench/ssd/65
posted by Lifeson at 7:01 AM on December 27, 2011
posted by Lifeson at 7:01 AM on December 27, 2011
Best answer: It's not as detailed as you have requested, but start looking at Tom's Hardware's Best SSDs for the Money - November. Presumably they'll post a December or January article soonish.
posted by Edogy at 7:16 AM on December 27, 2011 [1 favorite]
posted by Edogy at 7:16 AM on December 27, 2011 [1 favorite]
The SSD market works on the hot/crazy scale. Personally, anything that isn't intel looks like garbage to me. Sure, its a little faster, but your fail rate is much higher. OCZ is popular but fail prone. This is a contentious issue, but thus far I haven't had any Intels fail on me and my fail-rate with OCZ floats around 50%.
posted by damn dirty ape at 11:09 AM on December 27, 2011 [1 favorite]
posted by damn dirty ape at 11:09 AM on December 27, 2011 [1 favorite]
my fail-rate with OCZ floats around 50%.
damn dirty ape has a legitimate point here. The thing you need to keep an eye on is the controller chip on the SSD. Read as many reviews as you can on the SSD before buying. OCZ had a recent run of bad controller chips on many models of their SSDs. The current rumor is that the issue has passed, but it's put a black eye on them. Corsair had to recall the Force 3 SSDs. Intel SSDs have had fairly low failure rates, but there's been issues popping up that Intel 320's will lose significant capacity if a power failure occurs.
There are a few resources I would trust. The Best SSD article linked above will sometimes point out faulty drives. Hardware.fr (French-language) reports on return percentages. I also like to read the reviews on Newegg. Don't trust one single review, look for patterns.
posted by Mister Fabulous at 12:27 PM on December 27, 2011 [1 favorite]
damn dirty ape has a legitimate point here. The thing you need to keep an eye on is the controller chip on the SSD. Read as many reviews as you can on the SSD before buying. OCZ had a recent run of bad controller chips on many models of their SSDs. The current rumor is that the issue has passed, but it's put a black eye on them. Corsair had to recall the Force 3 SSDs. Intel SSDs have had fairly low failure rates, but there's been issues popping up that Intel 320's will lose significant capacity if a power failure occurs.
There are a few resources I would trust. The Best SSD article linked above will sometimes point out faulty drives. Hardware.fr (French-language) reports on return percentages. I also like to read the reviews on Newegg. Don't trust one single review, look for patterns.
posted by Mister Fabulous at 12:27 PM on December 27, 2011 [1 favorite]
Follow up to my earlier comment, here's Best SSDs For The Money: December 2011.
posted by Edogy at 11:53 AM on December 31, 2011
posted by Edogy at 11:53 AM on December 31, 2011
This thread is closed to new comments.
posted by tgrundke at 5:12 AM on December 27, 2011