UNSOLVED

DaGrin

updated

14 years ago

D

DaGrin

87 Posts

1

4597

June 12th, 2012 00:00

VNX pools and queue length

We're about to design a VNX5300 and are wondering if it's better to use just one pool or two seperate pools.

And can somebody tell me more about queue length of a lun in a pool.

Does a lun in a pool has a fixed queue length or is that relative to the number of drives in a pool.

Regards,

John

  • Madhug1

    2 Intern

    127 Posts

    1024

    0

    Posted June 12th, 2012 01:00

    Hi,

    Using one pool or two pools depends on various factors like type of applications, etc....

    Virtual Provisioning pools and traditional luns have same queue lengths, although with pools,the entries apply to the pools private RAID group and are fixed.

    For more information, you may refer EMC Unified Storage Best practices for performances and availability from Powerlink

    https://powerlink.emc.com/nsepn/webapps/btg548664833igtcuup4826/km/live1/en_US/Offering_Technical/White_Paper/h8268_VNX_Block_best_practices.pdf?

    Thanks

    Madhu

  • DaGrin

    87 Posts

    1024

    0

    Posted June 12th, 2012 01:00

    But does that mean that the more drives I have in my pool, the more private raidgroups, the bigger the queue is or is the queue per lun in a pool always fixed. So it doesn't matter how many drives I have in my pool it's always a certain depth.

  • ksnell

    38 Posts

    1024

    1

    Posted June 12th, 2012 06:00

    Hi John. Choice for 1 or more pools should be down to a few factors;

    Failure domains; as pools use a combination of private RAID Groups and Private LUNs, multiple drive failures within a private RG will render a pool invalid. The probablility is extremely low within each private RAID Group but more drives in a pool, more private RAID groups, actually lowers the calculated availability.

    Aggregate capacity increases with more drives; have to consider backup and recover strategy, etc.

    Performance domains relative to access characteristics. As pool LUNs will share the underlying disk resources, sharing random and sequential activity isn't necessarily a great idea. So if you have predominantly sequential workloads, best to seperate those. Of course, it's performance so it'll depend on requirements, concurrency of the workloads, how I/O gets distributed across the pool affected by working data ranges, etc.

    Queue - as indicated by Ganapa, the queue is 'currently' goverened by the preferred drive count modulous. What that really means is that a RAID5 pool with a modulous of 5 (preferred drive multiple for the pool), will dictate a queue of (14* efective data drives thus 4) +32 = 88.

    The next major release for Unified VNX has changes on the queue mechanism so check out the best practices guide that's in rewrite now for release aligned with that major update.

    Hope that answers your inquiry.

    Regards,

    ~Keith