back to article Analytics are better with shared storage and not scale-out, says IBM

IBM has a packaged IBM Data Engine for Analytics – Power Systems Edition, a cluster that includes the integrated servers, parallel file system storage, network switches, and software needed to run MapReduce-based workloads. It is described as a technical-computing architecture that supports running Big Data-related workloads …

  1. Anonymous Coward
    Anonymous Coward

    ESS distributes the data and parity information evenly across all disks in the system [which] allows for storage rebuilds triggered by a disk failure to complete faster than traditional RAID rebuilds by distributing the work across many disks rather than just a few.

    Unless probability theory has changed, the overall MTBF will be MTBF/N where N is the number of drives. Definitely problematic as an increase in N increases the time in rebuilding becomes ever closer to operational time ever increasing the probability of array rebuild failure. So, what additional magic have in there?

    1. Anonymous Coward
      Anonymous Coward

      Not Magic, Just Maths

      MTTF of a single drive is not key. If I have 10,000s of drives a whole bunch will fail. Double protection has been around for a while and triple is becoming more common. Failure domains and fast rebuild mechanism protect the data where traditional RAID would see data lost. There are a lot of products being sold that still use 90s techniques but the storage world is moving on (and it needs to).

      There are a range of vendors using FEC or similar to achieve much faster rebuild times compared to traditional RAID. Not magic, just maths ... do some research. Not just Spectrum but XIV, CleverSafe, Isilon et al all achieve faster than trad. RAID rebuild times.

      IBM seem to be following the Isilon data lake play here and the standard Hadoop is looking at moving to FEC vs. it's v1 triple mirror. A lot of change to come.

      1. Anonymous Coward
        Anonymous Coward

        Re: Not Magic, Just Maths

        here's the text from a comment in a The Register forums from last year - courtesy of "foo_bar_baz" - as far as I understand things, he nails the explanation....

        "Internet commenting is always entertaining, statements and declarations based on guesswork and conjecture. Let me indulge in some guesswork myself. Maybe the "rebuild" time is not about rebuilding a single disk, but about rebuilding the entire storage system from "degraded" to "healthy".

        Let's say your data is distributed across 100 storage nodes. Any one chunk of data in a "healthy" system is stored on n nodes where 100 > n > 2. One node dies, so the array is now degraded. To "rebuild" the system you just have to copy the chunks to free space on a sufficient number of nodes to satisfy the above requirement. Given that GPFS gets its high performance from storing files in small chunks across many nodes (not just 2 as in RAID1), it follows that rebuilding is also very fast. "Rebuilding" does not even necessarily have to involve replacing the broken node with a new one."

        Given that GPFS has been around as a product since 1998 (and has heritage back to "Tiger Shark" in 1993), I think it's unlikely that IBM is "following the Isilon data lake play" (with that company being founded in 2001) - although I see that kind of assumption a lot.... ;-)

  2. Convinient bug

    Re-inventing a need for centralized storage - all over again

    AFAIK, the whole point of HDFS was to be distributed, redundant, fault tolerant and working with cheap drives that fail. Now they're trying to convince us to go back to SAN? I don't think so. We already passed that river.

    1. Anonymous Coward
      Anonymous Coward

      Re: Re-inventing a need for centralized storage - all over again

      There is no SAN inside IBM's Elastic Storage Server (ESS) at all. GPFS Native Raid (GNR) is part of file system software layer and does manage JBODs which are connected via LSI SAS adapters. All hardware is COTS. The connection to the outside world is via 3x Mellanox 2-Port adapters per ESS node offering 10/40 GbEthernet or 56 Gb Infiniband. If you add ESS boxes to your installation you add capacity, bandwidth and more HA. If you like to add your own stuff e.g. running internal disks inside Linux nodes - no problem; mix and match is ok. If you don't like IBM Hardware - DDN, Seagate, Lenovo and others offer the same file system - mix and match is possible.

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

Other stories you might like

Biting the hand that feeds IT © 1998–2022