Over the past eighteen months, my job has migrated from a wide-ranging network engineering and systems administration role into a much more focused storage engineering role.  I've had the opportunity to work with some of the largest and fastest commercial storage devices on the market and I'm constantly barraged by amazing new storage and archiving technologies.  As you guys may have noticed, my home network tends to reflect my professional occupation, and it's time I started catching up on storage.

Over time, I've become increasingly aware of how ridiculously dated the storage system is in the Bobulator.  Although SSDs have been commercially available at the consumer/prosumer level since 2007, I've been avoiding them like the plague for a variety of reasons.  The initial models were ridiculously expensive, had incredibly low capacity (32GB was the largest size available), and had substantial issues with poor lifespans due to a lack of TRIM support.  I could save up money for a good SSD if the performance warranted it, but I can't handle a lack of capacity and I absolutely refuse to jeopardize the safety of my data with an unreliable drive.

As always happens with hardware, the situation has changed substantially over the past couple years.  SSD prices have dropped by 66% over two years from roughly $3.00/GB to roughly $1.00/GB.  Some decent manufacturers are as low as $0.85/GB now.  Capacity has also improved substantially.  64GB SSDs were the largest commonly available drives two years ago, and now quite a few 512GB SSDs are on the market (albeit still very pricy).

It took awhile to wade through the massive number of SSDs on the market to find the right one for the Bobulator.  Whereas hard drives require very strict tolerances to manufacture (thus requiring very expensive factories to build them), SSDs are a lot more forgiving and can be built in factories that were designed for other solid state components such as motherboards, video cards, etc.  As such, there are a veritable crapload of vendors hawking SSDs.  You are limited to Western Digital/Maxtor/Seagate/Hitachi for a spinning disk, but there are easily 30+ vendors selling SSDs.

Here were my requirements for my first SSD:

• At least 200GB capacity, with 256GB desired.
• High reliability ratings from consumers.
• Blazing fast throughput.  At least 400MBps random reads and writes, 40,000+ IOPS.

I finally settled on this bad boy:



256GB is the sweet spot for me.  This is only replacing my C: drive.  I keep all of my applications on a D: drive, mass storage on an E: drive, and Steam games on an F: drive.  I am currently consuming ~100GB on my C: drive, much of which is in my user profile.  Some day I'll move my user profile to another drive, but that is time consuming and I'm lazy.

The Samsung 840 Pro series claims up to 540MBps for sequential reads and 520MBps and up to 100,000 IOPS for random reads, which is ridiculously fast.  Real world performance testing from several different reviewers showed an average of 515MBps for sequential reads and 475MBps for sequential writes, which is insanely fast.  Most of the other comparable SSDs offer about half that performance.  Sadly, the SATA 6GB ports in my computer are attached to a crappy Marvell controller, so I'll likely only see about 2/3 of that.  I'm planning on rebuilding the Bobulator in about 12 months, though, so it's wise to plan ahead.


-b0b
(...more to come.)

So, here's what it comes down to: performance!  I knew the crappy Marvell 9128 controller on my motherboard would be a limiting factor, but I was still hoping for decent performance.  The Bobulator is now two years old and SATA 6GB was a brand new technology in December 2010.  It wasn't supported by Intel's ICH10R chipset, so Asus added SATA 6GB to their high-end motherboards by using a third-party chip.  These "tack-on" solutions always suck, but it's what I've got.  I'll build a new Bobulator in about 12 months, so I'll be able to unlock the total performance of this SSD when that happens.

I used a few different tools to test performance.  Most of the testing was done with a tool called IOMeter.  IOMeter is freeware and it's the industry-accepted tool for storage performance benchmarking.  It's the same tool I use for performance testing on multimillion dollar storage appliances, and it works just as well for standalone disks on gaming PCs.  ;p


Random IO
The first test was for random IO.  Random IO occurs when you have a whole bunch of files being accessed simultaneously, causing the drive to hop all over the place grabbing chunks of data.  A good example is the launching of an operating system which requires a bunch of files to be accessed at the same time.  SSDs are insanely good at random IO because they can access non-sequential sectors just as quickly as sequential sectors.  It takes the same amount of time to access sectors 1/32/179 as 1/2/3.  This isn't the case with spinning disks which can only read or write to a given sector once the correct portion of the platter is aligned underneath the head.

My existing hard drive was able to provide 380 IOPS (input/output operations per second, e.g. a single read or write operation) for random reads and 366 IOPS for random writes.  That's actually not bad for a 7,200RPM spinning disk.  They max out at approximately 80 IOPS natively and the rest is all caching, hence the value of a large cache (64MB or higher) on a spinning disk.  The SSD absolutely blew it out of the water, though, with nearly 57,000 IOPS for random reads and just over 46,000 IOPS write.  These lead to mind-boggling improvements of 14,945% and 12,593%, respectively.




Sequential IO
IOPS can be misleading, though, as the number of operations you can perform is meaningless if those operations don't reflect real-world usage.  Sequential IO is a much better test of true performance.  Unlike random IO, sequential IO is a measure of performance when sectors are read in order (e.g. 1/2/3 instead of 17/5/32).  A good example of sequential IO would be a straight reading of a file like an HD movie file.  Large numbers of sectors are read in numerical order to allow the media player to stream your movie from the disk drive.

My existing drive provided 46MBps for sequential reads and 59MBps for sequential writes.  These numbers are a bit backward, as read operations should always outpace write operations.  I chalk this up to an overtaxed drive.  I had three volumes on a single physical disk (C: for OS, D: for applications, and E: for mass storage), so I was splitting the paltry IO across three different volumes.  For instance, any IO used by the OS on drive C: would be unavailable for applications on drive E:.  By moving the C: drive to the SSD, I was not only accelerating the C: drive substantially but also reducing the IO constraints on the D: and E: drives. 

The SSD provided 383MBps for sequential reads and 238MBps for sequential writes.  That's an improvement of 832.61% and 403.39%, respectively.  Sadly, this is where the weakness of the Marvell chip really shows up.  I should be seeing another 140-160MBps for reads and nearly 280MBps for sequential writes, but the Marvel 9128 just can't handle it.  I'm considering purchasing a high-end RAID adapter just to get the performance I'm paying for, but that money would can probably be put to better use by saving up for the next Bobulator.




Launch Times

For me, launch time isn't a huge concern as I very rarely reboot my system.  Like most of you, my PC runs 24/7.  Screw the environment!  With that said, I understand that many people use OS launch times as their benchmark of SSDs.  Keep in mind that I have a very "heavy" OS that loads a lot of extremely resource-intensive applications (e.g. Visual Studio, SQL Server, etc), so my load times won't be anywhere near as low as a system that is specifically engineered for fast boots.

Since my system is on an Active Directory domain, I have to provide credentials to log into the system.  As such, I'm breaking my results into two categories.  The first, Windows Launch, measures the time from hitting the power button to seeing the "Press Ctrl+Alt+Delete to Log On".  The second, Desktop Ready, measures the time from me hitting "enter" after entering my credentials to the point where everything is completely loaded and my primary disk backs down from 100% utilization.

The launch time for Windows was 65 seconds with the spinning disk.  This dropped to a mere 18 seconds after migrating to the SSD for a 361% improvement in boot time.  The desktop ready time was a very substantial 386 seconds on the spinning disk, or nearly six and a half minutes.  This is why you should never, ever put your applications and operating system on the same spinning disk!  The queue lengths were absolutely astronomical for that poor disk.  By improving the IO of the operating system and removing that burden from the disk containing my applications, I now have a ridiculously fast desktop ready time of 15 seconds.  That's a 2,526% improvement!




Final Thoughts

Overall, I'm incredibly satisfied with the Samsung 840 Pro series, and I would recommend it to any enthusiast that's looking to improve the performance of their system.  Honestly, I can't think of a single component I've ever replaced that had even half the impact of this SSD.  Just look at these numbers!




-b0b
(...sorry about the lengthy posts!)

I'm with ya, Stewie.  I have this odd problem where I want the best-of-the-best-of-the-best when it comes to hardware, so I'll suffer with crap hardware for way longer than I need to instead of compromising.

I'm planning on buying another identical drive in the next few weeks to move my D: drive (applications) to.  That will live my E: drive (mass storage) on it's own disk much like my current F: drive (Steam games).  It will also leave me enough space on the D: drive that I can install a handful of my current favorite Steam titles.  The new Steam client allows you to specify which drive you want to install a game to, so you don't have to mess around with junction points and other crappy hacks.


-b0b
(...can't wait to play a game off an SSD.)

I just purchased the 128GB version of the SSD you installed. Currently, I'm uninstalling a bunch of development tools to get my source disk size small enough to transfer. I don't have a windows disk or key, otherwise a new OS install would have been my approach as well.

Looking forward to getting my 5+ minute boot time down.

Oh yeah, tested it out yesterday and it's the best upgrade I've experienced for my machine. System restore was eating up a ton of storage on my source drive. I lowered the allocation percentage freeing up 50GB of space and allowing the transfer.

During the transfer I wasn't able to get above 55MB/s. Comparing that to your speeds, I must have had a really slow primary disk read.

I was telling a friend of mine who was building on a budget that if there was one thing he shouldn't skimp on is a primary SSD for at least his OS.

I think the only other time I've been this amazed with an upgrade it hardware is when I bought my first one gig hard drive for $100 all those years ago.

Sometimes our tech is so amazing it just shocks me.

X

Stick wrote on Apr 10^th, 2013 at 3:43pm:



I still use my very first thumb drive, bought it in 2001.  128MB, USB 1.0, and roughly $300.  Not my smartest purchase, but I'll get my money's worth out of it if it kills me.


-b0b
(...!)

Check out this amazingly compact 5MB hard drive:



They don't make 'em like they used to.  I work with similarly-sized (physically) storage appliances.  They hold about 5PB, or approximately 1 billion times as much data.  They also do all kinds of crazy stuff like deduplication, caching, archiving, cloning, etc. as opposed to simple storage.

This is one of my favorite comparison images:



The drive in the back is a 60MB IBM disk designed for PDP-10 servers.  When they were originally built, they cost over $25,000 each.  As you can see, the disk platters were about a half-inch thick and were roughly the size of vinyl records.  Each of the six platters held a whopping 10MB.

These days, high-end systems have 128GB of cache on a class 10 SD card, and they only cost about $150.  1TB enterprise SSDs have started to ship that are 2.5" in size, and 4TB spinning disks have been shipping in the same form factor for a couple years now.


-b0b
(...can always use more storage.)

Remember when we use to complain about the Window boot up time?  Now I think...it's not even worth getting a snack because the computer will be up.

It's amazing 3D configuration adds that much power to SSDs.

The_Fat_Man wrote on Aug 11^th, 2014 at 7:27pm:


Damn kids, get off my lawn!

It looks like the issues with the Samsung 840 EVO weren't entirely fixed by Samsung back in November.  I've been seeing really flaky performance off the disk again.  I ran another HDTach benchmark and performance is even worse than it was before I ran the performance restoration tool.



Average read speed went from 23MBps before the tool to 489MBps afterward, and is now down to an abysmal 12.6MBps.  Latency was at 1.2ms before the restoration, 0.1ms afterward, and has now skyrocketed to 2.6ms.

I'm tired of screwing with this drive, so I've ordered a Samsung 850 Pro 1TB drive to replace it.


-b0b
(...yay 3D NAND!)

I use several tools for benchmarking, but the simple one with the nifty graphs shown above is HDTach.  It's definitely worth a download.

The performance restoration application is a proprietary Samsung application created specifically to address the issue with the 840 EVO SSDs.  Once the firmware has been updated to a version that purportedly fixes the underlying bug that caused the slowdown, the performance restoration tool reads every single file on the SSD and rewrites it back to the drive to fix the issue.  Unfortunately, the performance restoration tool will only run one time and won't let you run it again, and the firmware fix obviously didn't correct the degradation of performance over time.


-b0b
(...is getting ready to swap to the 850 Pro now.)