Which SSD is best for backtesting

Discussion in 'Hardware' started by ideabox, Aug 4, 2011.

  1. ideabox


    All the reviews that are out have tested solid state drives for their random 4k read, random 4k write, sequential read and sequential write speeds as well as the overall speed when running common tasks.

    Which one of the above parameters is used during backtesting?

    Is it random 4k read, random 4k write, sequential read or sequential write?

    Trying to decide if the new Intel 510 series is better and faster than the X=25M G2 when it comes to backtesting on a NinjaTrader platform?

    It appears that the G2 is faster than the 510 series in 4k random reads and writes as can be seen by the 2nd and 3rd tables here: http://www.anandtech.com/show/4202/...sd-510-review/4

    Also will it matter if one uses a shorter time frame to backtest - like say 6 months or even 3 months?

  2. dejavu8


  3. ideabox


    It has been 6 months and no one has yet answered the questions in my opening post. I know there are real tech people with solid understanding of this here on ET. I was hoping that at least a couple of people would know this.
  4. Here is an excellent white paper on SLC vs. MLC technology. It truly is a "get what you pay for" difference (although there still is a massive difference in prices between SLC and MLC).

    http://www.supertalent.com/datasheets/SLC_vs_MLC whitepaper.pdf

    The answer is "it depends on your backtest engine".

    If you have a massive (local) tick database you may be wanting read speeds but if you have a massive external tick database (pulling from your broker for example) then you will use very little HDD and want network speeds.

    Prior to SSDs being affordable, people looked for the largest amount of cache in a "spinny" HDD (64mb) and used a decent RAID card with a ton of onboard system RAM. You would take several days worth of tick data and load it into memory. The process of backtesting was slower than the transfer of data into memory - but still disk read only since it read the data on disk, loaded into memory and then purged from memory as it was used.

    This is only needed when backtest engine can process faster than your disk read speed. If your backtest engine is slower than your disk read/write you don't need to do anything as the disks are not the bottleneck.

    Only you can answer this question but I'd say that for 99.999% of the world a SSD will make very little difference since the data tends to generally be so large, you will spend thousands of dollars on a decent SSD setup with RAID card that can handle SSDs - when you are ready to spend that kind of money to get true RAID'ed performance from SSDs you'll know what you need.

    What platform are you on? How large is your tick database? Where is your tick database?

  5. How presumptuous and demanding.

    I could assist you, but after that remark you've made I won't do it for free.

    Try giving back to the community first.. we're not your personal free IT staff.
  6. ideabox


    Winston, thanks - that was informative.

    I'll be using NinjaTrader. I am not sure if I'll be using tick data - if I will it will be about 3 years in timespan. I was planning on using 1 minute data going back to at least 2000.

    Data will be stored on the SSD.

    With this in mind which of the five elements are most important?

    random 4k read,
    random 4k write,
    sequential read,
    sequential write,
    cache size

  7. The name of the game is the most read and write speeds per second. Usually all specs go up with read and write.

    The more money you spend, the better.

    You'll have a difficult time with anything under 120gb if you have any software at all.

    Intel SSD's are very expensive. Most 'enthusiastic' use Patriot, OCZ etc.
  8. Cuz everyone is doing it in a virtual machine OS with 32gb
    Everything in memory dude.

  9. All garbage. Serious work requires a Fiber channel card connected to a TMS Ramsan-720
    #10     Feb 13, 2012