View a markdown version of this page

Storage types - Best Practices for Deploying SAS Server on AWS
Permanent SAS data storage

This whitepaper is for historical reference only. Some content might be outdated and some links might not be available.

Storage types

AWS has many storage types for temporary and permanent requirements. In this section, we address the options for field experiences and lab testing with the SAS on AWS storage options.

Permanent SAS data storage

Permanent SAS storage is used for SAS 9.4, SAS data files, and SAS Viya CAS tables. SAS 9.4 data files hold either a SAS dataset holding actual data, or a SAS non- materialized view definition that references data stored elsewhere.

Viya CAS is not only an analytic and transformation engine, it is also a data server. It loads data into a CAS table in order to analyze and process. The format of these tables

can vary including SASHDAT, CSV, Oracle, SQL Server, or Hadoop, and it is backed into a permanent storage with content stored in-memory.

The following permanent storage options are suggested to support the SAS data files and SAS Viya CAS tables:

  • Elastic Block Storage (EBS) – Stripe together a minimum of 4 EBS volumes for I/O bandwidth aggregation.

    • EBS ST1 (throughput optimized HDD) Storage designed for large block sequential I/O. A 12.5 TB volume can sustain 500 MB/second. If the volume size is less than 500 MB per second of total bandwidth, it can be observed during the burst window.

    • For high-throughput read-heavy workloads (like in SAS), update the read-ahead setting on EBS ST1/IO1 from default 256 KB to 8 MB. EBS IO1 (provisioned IOS SSD) storage can also be used.

    • Other EBS storage types like GP2 (general purpose storage) and SC1 (cold storage) are not suitable for permanent SAS 9 or SAS CAS data files.

    • RAID 0 configuration is preferential because fault tolerance is not a determining criterion for these workloads.

    • Customers can also choose to have EBS IO1 volumes (provisioned storage). However, costs would increase as IO1 volumes are charged by storage and by provisioned IOPS. For ex – 32K IOPS can yield as much as 500 MB/sec but customers would pay an additional amount for the desired provisioned IOPS.

    • Using SAS/Access to Redshift, SAS Datasets can be loaded into Amazon S3/Amazon Redshift using Amazon S3 capabilities of multi-part upload, transfer acceleration, and COPY/UNLOAD to Amazon Redshift for relational storage.

Small diagram that shows directional bulk and standard loading from SAS into Amazon S3 or Amazon Redshift.

SAS - Redshift Bulk Load

Temporary SAS data storage

Most temporary SAS storages used in SAS WORK, SAS UTILLOC, and CAS_DISK_CACHE will not persist through reboots and are considered ephemeral storage.

  • I3 instances feature low latency NVMe SSDs striped together with RAID0. Use NVMe devices to support high bandwidth, low latency, and sequential I/O .

  • If additional storage is required, default to permanent SAS storage.