back to article This flash is too slow. This DRAM is too small. This bowl of data is just right. It's SCM flavored

Among all the uncertainties in this world right now, there is at least one constant: applications still absolutely rely on data storage. The performance of code – from enterprise resource planning (ERP) and online transactional processing (OLTP) software to machine-learning tools, high-performance computing workloads, and real …

  1. Duncan Macdonald Silver badge

    Not enough memory channels

    The majority of CPUs do not have sufficient memory channels to make good use of SCM. Without being able to dedicate a memory channel to SCM, it is relegated to fast NVMe storage. (The AMD EPYC range is one of the few that can afford to sacrifice memory channels to SCM. One SCM memory channel would reduce the maximum DRAM on EPYC from 2 TB to 1.75TB per CPU.)

    Another problem with SCM is that it is only effective for local storage (SCM and CPUs on a single motherboard) - serving the data on SCM over any sort of network negates its advantages. For a cloud server this implies that jobs using SCM cannot be easily migrated to another node.

    1. DCFusor Silver badge

      Re: Not enough memory channels

      Yeah, but.

      If you liked the Beatles on 45's, you're going to love them on vinyl.

      We'll even sell them to you on 8 track tapes for your car!

      Oh, now we have cassettes! Better, new, shiny!

      And wow, you're really going to adore them on CD!

      (but we stopped biting and paying over and over after that - so DVD etc profit).

      Maybe we'll just buy ram...

  2. Anonymous Coward
    Anonymous Coward

    This is ridiculous vendor pimping nonsense

    SCM on a storage array is simply another tier of storage... ON AN ARRAY! You don't solve the latency problem this way. It's literally lipstick on a pig. A big expensive storage pig.

    If you want maximum application performance, you need that Persistent Memory Storage ON THE SERVER and not as a tier on the array. Single digits latencies! Optane delivers this today.

    Why does this seem to be so difficult for industry analysts to understand?

    1. Rob Isrob

      Re: This is ridiculous vendor pimping nonsense

      I don't know... I'll take a stab at it. First up is cost. Keep in mind "hot" data makes up less than 10% of data. So in large Enterprise environments with 300, 1000 servers impractical to stick optane on the ones that "need" it. Secondly, latency over NVMe is less than 40 us in shared storage. Not shabby. Then the whole aspect of failuire domain. You move that out of Enterprise SAN to multiple hosts.... congrats, you just increased management and points of failuire. There are probably some of the reasons Pure and others realize optane is good Enterprise SAN fit.

    2. Stuart 18

      Re: This is ridiculous vendor pimping nonsense

      --Optane delivers this today.---

      Umm yeah RTA Optane is SCM.

  3. Dave 13


    Array-centric thinking is great.. but what if you actually stuff SCM into HCI nodes connected with 100Bbe or IB.. and lean on scale? I'm talking to you, ScaleIO (now VxFlex OS).

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019