SSD 102: The Ins And Outs Of Solid State Storage

Performance Limits And Pitfalls

Let’s cover a few things that must be considered when planning a transition to SSDs. We've already mentioned issues with sustained performance in prior SSD articles. This was an issue with older drive generations but has been improving over time. Still, you should take precautions. Most of the following items are valid for client PCs as well as for business and enterprise deployments.

RAID Support

Technically, all SATA drives can be utilized in RAID arrays using suitable controllers. However, some products don't deliver expected performance in RAID configurations. This is often a firmware issue, but should be checked nonetheless before investing in specific SSD products. Sometimes the TRIM feature (see below) won’t work on RAID configurations due to incomplete AHCI support.

Defragmentation

Hard drives should be defragmented on a regular basis to make sure that frequently-used files are available on the faster sectors and that larger files aren’t unnecessarily scattered across the hard drive, which increases the time needed to retrieve them. Hard drives need to read and write sequentially as often as possible because head physical movements introduce latency. This isn't a problem for SSDs since data is always distributed by the SSD controller all over the drive. This is normal. Applying defragmentation tools to SSDs will be counter-productive, wearing down flash cells without introducing any advantage. Hence you should switch defrag off.

TRIM Support

The TRIM feature is used by the operating system to inform the SSDs about blocks that were deleted and are available for new data. This allows SSDs to optimize writes in a way that maintains maximum write performance and optimizes wear leveling. SSDs do this internally, but the process is more efficient when triggered by the file-aware operating system. TRIM requires support in both the SSD’s firmware and the operating system. Drivers and the storage controller must support AHCI, as well. This often means using Windows 7 or Windows Server 2008 R2, Linux kernel 2.6.33, or FreeBSD 9. All older system versions will not support TRIM, which can result in severely limited write performance. However, TRIM support is complex, so double check that support is possible and enabled.

TRIM is supported by Intel SSDs with firmware 02HA and up, Indilinx 1819 and up, all JMicron 612 and 618 devices, Marvell hardware with firmware 0002, all SandForce devices, Samsung’s VBM19C1Q firmware for PB22 drives, and all Samsung 470 SSDs. The tool Crystal Disk Info provides great help with firmware and SSD checking.

Create a new thread in the US Reviews comments forum about this subject
This thread is closed for comments
26 comments
    Your comment
  • LaloFG
    Very good article; times of most affordable capacity in SSD units is coming...
    2
  • Lewis57
    A very good article. I love these articles explaining everything. I'm planning on buying two OCZ Vertex 2E 60GB for RAID-0 when I get enough money. Can't wait, should be one hell of an upgrade from a single 5400rpm WD green drive.
    1
  • ares1214
    Memristors might make SSD's sorter lived than people thought, but who knows. Great article btw.
    0
  • JoeSchmuck
    From what I understand, TRIM is supported under IDE mode using Win7 as well so you do not need AHCI. I have a Samsung’s VBM19C1Q firmware device and running IDE mode.
    0
  • hemburger
    Great article!! :D
    0
  • Anonymous
    Earlier this year we deployed a 5 node failover cluster with iSCSI backend. Each of the VM Host servers utilize a pair of solid state drives for booting and operating, with VM's running off of iSCSI shared cluster volumes. The servers are unbelievably fast and stable - 6 months of 100% uptime on Windows 2008R2. We only use magnetic HDD's now for transporting backups off site.
    0
  • Anonymous
    One thing that I'm very curious, if we follow Tomshardware's advice to turn off disk defragmentation, the files on SSD would be defragmented over time.

    Upon SSD data loss, can we recover the data files if it's defragmented, especially on a SSD that has never been defragmented as Tomshardware had recommended?
    0
  • randomizer
    Defragmentation of an SSD is not entirely unnecessary. It's important to distinguish between file fragmentation and free space fragmentation. The former is not an issue with SSDs because all parts of an SSD can be read at the same rate (the same is true for writing if the blocks are clean). But fragmentation of free space, whereby free space is largely distributed across partially-filled blocks, can severely reduce the performance of an SSD. Any time a file of <512kB is written to an SSD, it will take up only part of a block. However, the SSD will eventually run out of clean blocks and will need to re-arrange the data by erasing partially-filled blocks and consolidating them to free up more blocks for further writing. Running a free space defragmentation on the drive will aggressively consolidate the data on-demand so that you don't have the problem occurring when you didn't plan for it.

    Most SSDs will perform this process themselves when idle for extended periods, but it happens at a slow rate. This is what most manufacturers refer to when they talk about Garbage Collection.
    0
  • Alvin Smith
    Please send me the four fastest 256GB SSDs on the market, so that I might perform my own comparison ... I'll just sit by the door and wait for UPS to arrive.

    Thanks, in advance !!

    = Alvin =
    0
  • gordonaus
    I put an SSD in my new computer and it was good but after i got the firmware update and changed to AHCI it was AMAZING (OZC Vertex 2 60gb). I would say tho that 60 gb is not enough, i installed windows photoshop and a few other design programs and i only have 20GB left
    0
  • compton
    Great article. I'm happy to see it in the mix. I'm sticking with my Intel x25-V and OCZ Agility 60 for a little while, but who knows what the future will bring.
    0
  • cadder
    I haven't seen this addressed before- how does the reliability and data safety of an SSD compare to a mechanical hard drive?
    0
  • Anonymous
    It is much cheaper to buy 2 Hitachi 7200K disks which are quite reliable and compact RAID enclosure like Jou Jye. And you can have same SSD performance with 1TB drive for the price of one 256GB SSD. I have to mention that I can have up to 2,5Gbps max transfer rate which is not far from SATA II limit.
    I am using same configuration on desktop. What I have noticed is that performance is actually much better than I expected. That is probably because of cache memory. If you have drives with big cache then in RAID stripe configuration those caches logically combine. In case of good desktop drive you can easily have 64MB cache. BTW I looked at the SSD drives caches - wow I know where performance comes from. :) Actually not from SSD technology as such.

    I think SSD is overrated right now. They have to be 4x cheaper. Otherwise it makes no sense. Next year they will be 2x cheaper and after one more year they will 2x more cheaper. So actually technology still needs two years to be usable.

    My recommendation: stick to SATA and RAID - save the money. If you need little storage and max comfort then use SSD.
    -2
  • d1rtyju1c3
    This was a great article. I learned alot of new thing I never knew about SSD's. Now I will be able to make a better selection when I add a SSD.
    0
  • Keeper
    dvdeo,

    You save a lot of money with SSDs, simply because their watt consumption is really low. So, in long term (say 1y) you will be saving enough money to probably buy those Hitachi 7200K for free.

    Energy efficiency is the key factor with SSDs.
    0
  • JoeSchmuck
    I understand data reliability for SSD cells that may have reached thier maximum writing number, that the cells can still be read making the data at least available which is much better than a mechanical hard drive since when they fail, ut's usually not good.
    0
  • bitterman0
    Keeper... So, in long term (say 1y) you will be saving enough money to probably buy those Hitachi 7200K for free...

    The power consumption difference of a single drive is negligible for the purposes of generating any tangible savings on the electric bill. Let's assume the average power consumption difference between HDD and SSD is 5W, and the system that employs the drive is up 24/7/365. Also, let's assume that your electricity cost is 14 cents per kWh (that's what I'm paying on average, your mileage may vary). Thus 0.005kW * 24h * 365d * $0.14 = $6.132 - that's your annual savings (to be clear, that's six dollars and some change, not six thousand). Surely, if you employ hundreds upon hundreds of drives, the savings will add up, but in the end the up-front investment into SSD's higher cost is not likely to pay off within the SSD lifetime, not to mention to get you any savings.

    On a separate note, I do believe that longevity of drives is one of the major factors that affects the purchase decision. For enterprise use, if the drive is constantly hammered by writes (say, a database file is stored on it), the rate of wearing out re-writable flash is likely to be higher than the rate of failure of magnetic drives (certain 10K RPM IDE drives notwithstanding).

    ... if only SSD were more affordable! But, perhaps, the rumored adoption of 2Xnm technology for NAND by Intel by the end of this year will finally put enough pressure on the market to bring down prices to the realm of affordability. One can only hope.
    0
  • doorspawn
    Can someone shed some light on a query I'm sure many of us have here:

    Why is the block size so large?
    What makes a 4KB or even 256B block a bad idea?
    Is it there's a large per-block component that can't be shrunk?
    Is it that blocks need to be insulated from each other so that high-voltage instructions (perhaps clear) don't leak?
    Those are purely guesses.
    0
  • Anonymous
    Good overview article, error on last graph:
    5.5 watts to 1.7 watts is not "1/3 Reduced" as per label - it is "2/3 reduced" or "Reduced to a 1/3"
    0
  • dEAne
    thanks for this article.
    0