For the last time – Raid-S does not mean “STRIPED”

RAID-S isn’t striped.  If it were it would be Raid-5.

EMC utilized Raid-S as a stop-gap to the fact that they didn’t offer Raid5. (the constant XOR computations slowed down performance – until the faster DMX line came up which now allows for both 3+1 and 7+1 raid5.

Raid-S takes three volumesand computes a parity bit based on them, but lets them maintain their separate identities.

so SYMDEV1, SYMDEV2, and SYMDEV3 each have data, and can be mounted as /u01, /u02, and /u03. If a volumeis lost, that volume’s data is extrapolated from the parity volume. If two volume are lost (either two data volumes or one data volume and one parity volume) the data on the third (or remaining two in the event of a parity volume failure)

This actually gives you an added level of protection in that you don’t lose the entire group on the failure of multiple volumes as you do in Raid5.

Note I said volumes and not disks. Raid-S volumes are hypervolumes (or partitions) of disks and not whole physical disks.

9 comments

Skip to comment form

    • on April 7, 2007 at 2:03 pm
    • Reply

    It is funny how definitions of RAID are changed. I was working on an installation this week that involved an HP DL320s NAS box (12x250GB SATA drives).
    According to the HP SMart Array RAID card the first 2 drives were set as RAID 1+0 for the OS (and then partioned to 30GB with all the rest of the space wasted – but that’s another moan).
    Now, correct me if I’m wrong, but isn’t RAID 1+0 involve at least 4 drives.
    I’m still trying to figure out how they think this is RAID1+0??

  1. You’re right, RAID1+0 requires a minimum of 4 drives. Depending on how you write it, it’s “Mirrored then Striped (1+0) or “Striped then Mirrored” (0+1)

    I’ve heard that 1+0 actually manages to be faster than (0+1), probably owing to the priority of whether you’re writing to the strip first or the mirror first. (Write performance is typically better writing to a Raid0 set, because parity doesn’t need to be computed) If that write is completed before the mirroring (or independant of the mirroring) your performance will be faster. If the mirror of the write has to complete before the ACK is sent back to the host, you’ve given up any performance gain you might have received.

    As a rule we have the following:

    Raid-0 – Striping w/o parity
    Raid-1 – Straight Mirroring
    Raid-3 – Striping with a dedicated parity drive
    Raid-5 – Striping with distributed parity
    Raid-6 – Striping with a mirrored, dedicated parity drive.
    Raid10 – Striping first then mirroring the stripe.

    Then Raid-S – Whole volume parity calc – no striping. (EMC)

    There are many other implementations, these are the ones most commonly seen.

    • on April 7, 2007 at 7:38 pm
    • Reply

    Yeah, my guess is that it defaults to RAID 1 when 1+0 is selected across only two drives. Which still seems like false advertising to me when you query the RAID card. 🙂

  2. Well, I guess you could call a single disk a “stripe of one” but I don’t think that really is how it is.

    I had my SATA set in my workstation running at RAID-1/0 and performance SUCKED, but changing it to RAID5 seems to have solved the performance issues.

    When you see that, don’t just assume that Raid1/0 sucks, but simply that the people who wrote the firmware for the controller didn’t understand what order to write the data in.

    Also – most controllers need to be manually placed in “cached IO mode” otherwise they just do plan raw disk writes and you don’t see any real benefit.

    However in cached IO mode, make sure you’ve got a battery backup on the unit.

    • on April 21, 2007 at 4:57 am
    • Reply

    Ah yes, our old friend firmware. Back in my hard drive days I was working with a firmware guy at with a RAID card from one of the ‘Taiwan Inc’ companies that they were using in one of there systems. His comment on it’s firmware was “to call it a piece of would be an insult to the piece of ” :-)v

  3. Yeah, I’m not as thrilled with this motherboard as I think I should be. Though it’s an intel SATA controller, so a firmware upgrade should be easily doable, right?

    In theory that is.

    • on August 10, 2007 at 4:33 pm
    • Reply

    Dear SAN gurus…

    Kindly lemm eknow the command to check the free LUNS available out of the total LUNs in a symmetrix/DMX 3000 attached to a SUN host…wd be thankful to ya…i think it might be one of the symconfigure command..plz help..

    Thanks n reg
    ~Dee

    • Robert on October 23, 2011 at 4:59 pm
    • Reply

    Thank you for this post, but let me know, if a system running this EMC file system, got broken:
    – how do you do to recover the data?

    Thank you for the hint (if any)

    Robert

    1. Sadly, you don’t. Hence the need for good backups regardless of what kind of storage you use.

Leave a Reply

Your email address will not be published.