Primary Storage

 View Only
  • 1.  FCM2/FCM3 review

    Posted Wed June 08, 2022 03:47 AM
    Hi guys,

     Is there anyone with FCM2/FCM3 in production ? Do you have any real-life experience you want to share ?  I know the marketing stuff around the FlashCore Modules, I also checked the presentation done at Flash Memory Summit by IBM but I want to see if they are really a better choice than TLC drives. I need to build a new flash pool in the near future using FS5200 or FS7300 for heavy SQL workloads.  What are the FCM latencies vs TLC latencies ? What about the IOPS ? What about the endurance ?  What happens when they are getting close to 80% usage (the QLC drives are usually very slow when they start to fill) ? 


    Adrian Nicolae

  • 2.  RE: FCM2/FCM3 review

    IBM Champion
    Posted Thu June 09, 2022 05:12 AM
    Edited by Nezih Boyacioglu Thu June 09, 2022 05:18 AM
    FCM2 and FCM3 same SLC/QLC technology. FCM's uses Smart Data Placement with dynamic SLC allocation. FCM performance better than common TLC.
    Spectrum Virtualize v8.5 and FCM 3 "Hints" improve performance and reduce latency. For IOPS, latency on your workload type and all other questions please contact your local IBM representative. They will help you to understand your current workload and simulate how 5200 or 7300 performs under your heavy sql workload. 

    Nezih Boyacioglu

  • 3.  RE: FCM2/FCM3 review

    Posted Fri June 10, 2022 09:59 AM
    The other major advantage to FCMs is the inline at speed hardware compression.  If the data that you will be storing on the systems is compressible, the FCMs offer data compression on ingress, that once compressed, is also how all data management on the back end occurs.  This helps to significantly lower the write amplification as compared to drives that do not have compressed data stored.  And using FCM compression, means that the performance over using other compression methods is overall better.  FCM3 increased the usable to effective ratio over FCM2 so this advantage can be even more realized, for data sets that can take advantage of it.

    Brent Yardley