Mainframe Storage

 View Only

Configuring the DS8880 - Using Small Extents

By Nick Clayton posted Tue February 12, 2019 03:21 PM

R8.1 of the DS8880 introduced a new configuration option. This provides the ability to configure a storage pool as either large or small extents. The small extent size is significantly smaller than the previous 1GB(FB)/ 1113cylinder (CKD) large extent size being 16MB (FB) / 21 cylinders (CKD).

Using small extents should now be considered the default when setting up a new DS8880 unless there are very specific reasons not to configure the DS8000 in this way. They provide a range of benefits both functionally and for performance. There is also no easy way of switching a DS8000 from large extents to small extents, so in most cases this decision will be for the life of the storage system.

The significantly increased granularity of allocation with small extents provides a much more efficient Thin Provisioning capability than large extents. If there is any intention to use thin provisioning either immediately or in the future then small extents should be used. The DS8880 also provides a capability to prevent overprovisioning using the OverProvisioning Ratio Limit and so it is possible to start using thin provisioning without initially overprovisioning devices and for this to be enforced by the storage system. As fully provisioning devices cannot be converted in place to thin provisioned the Overprovisioning Ratio Limit provides a stepping stone to later exploitation of thin provisioning for capacity savings.

The benefit of small extents is especially important when Thin Provisioned devices are used for Space Efficient FlashCopy or Safeguarded Copy Backup Capacity. It is possible to define a separate storage pool for thinly provisioned FlashCopy or Safeguarded Copy capacity but this results in an imbalance in the workload over the arrays and so using a common pool for sources and targets is recommended.

Small extents also provides a much more granular striping of data over the arrays which will spread sequential read or write bursts over a larger number of arrays than with large extents. Unless intending to use single rank extent pools and having a very balanced sequential write workload this will provide better overall performance than large extents despite large extents having ~15% higher throughput for a single array.

Small extents also improves the ability of Easy Tier to manage the hot and cold data especially when combined with thin provisioning. The DS8000 does not manage everything at the small extent level as this would be impractical but cold data is tracked individually on each small extent. Also if thin provisioned we do not have to manage the data that does not exist.

When using small extents this does introduce some limits for physical and logical capacity on the DS8880 depending on the amount of cache (512GB or lower and 1TB or higher). The limits are shown in the table below

Note : If using Easy Tier then the logical volume capacity above 2024TiB should only be used for Safeguarded Copy.

The important message here is that on the DS8886 having a 1TB cache gives you better physical AND logical scalability so especially if you are looking at thin provisioned FlashCopy or Safeguarded Copy with significant overprovisioning you want to carefully consider 1TB cache even if you don't need it from an application performance perspective

Other other important item to be aware of when exploiting thin provisioned devices with small extents is the behaviour of what we call Global Metadata. There are three copies of Global Metadata which are configured for maximum availability and so will be spread over three RAID adapter pairs unless the DS8000 has less than this number. This means for a Hybrid DS8880 unless the configuration has three HPFE Enclosure Pairs at least one copy of the Global Metadata will be on HDD. There is no specific capacity requirement for the Flash arrays as the Global Metadata is very small in capacity terms.

Global Metadata describes the configuration of the DS8000 including the information about the location of each extent and so needs to be updated when extent are allocated. In intensive write workload environments this can result in significant peaks of activity on the arrays holding the Global Metadata and so for optimal performance three HPFE Enclosure Pairs should be configured if possible.