EBS volume types: gp3 vs io2 vs st1 vs sc1

Try 3 questions on this

Pick an answer to see why each option is right or wrong.

1. A trading application needs a single volume delivering 150,000 IOPS at consistently sub-millisecond latency, and it will run on a Nitro-based instance. A solutions architect must choose between General Purpose SSD gp3 with its maximum settings and Provisioned IOPS SSD io2 Block Express. Which approach will meet these requirements?

   Right. In the game that’s Renown for your Design High-Performing Architectures building.

   Not that one: C is right. In the game your Design High-Performing Architectures building would lose a little Renown, and this question would come back later.

   Your answerA.Use gp3 at its maximum settings, because gp3 and io2 deliver the same IOPS on Nitro-based instances.A Nitro-based instance can drive up to 256,000 IOPS, but a gp3 volume cannot be provisioned beyond 80,000. The instance's capability does not raise the volume type's maximum.Your answerB.Use two gp3 volumes in a RAID 0 stripe, because striping is the supported way to get past one volume's IOPS limit.Striping aggregates IOPS across volumes and does nothing for per-I/O latency, which is the half of the requirement io2 Block Express is designed for.The answerC.Use io2 Block Express, because one gp3 volume can be provisioned to at most 80,000 IOPS.Both SSD types are plausible until the numbers are read, which is the deciding behaviour here. The table caps gp3 at 80,000 maximum IOPS and 2,000 MiB/s, and io2 Block Express at 256,000 IOPS and 4,000 MiB/s. 150,000 IOPS is therefore outside gp3's envelope and inside io2's.Your answerD.Use io1, because Provisioned IOPS SSD is a single family and io1 reaches the same ceiling as io2.io1's maximum is 64,000 IOPS against io2 Block Express's 256,000. Same family, different ceilings, and io1 is below the requirement.

   Show the full explanation and doc quotes

   The latency requirement points the same way: io2 Block Express is listed for "consistent sub-millisecond latency with average latency under 500 microseconds", and the footnote states it is "designed to deliver an average latency of under 500 microseconds for 16KiB I/O operations." The Nitro detail in the stem matters for a different reason - the footnote records that only Nitro-based instances support volumes provisioned above 64,000 IOPS.

   Written from docs.aws.amazon.com/ebs/latest/userguide/ebs-volume-types.html

2. A clustered application needs one volume attached to several EC2 instances at the same time, with the cluster software coordinating writes. A solutions architect must pick a volume type that supports this and must tell the team which types do not. Which solution meets these requirements?

   Right. In the game that’s Renown for your Design High-Performing Architectures building.

   Not that one: B is right. In the game your Design High-Performing Architectures building would lose a little Renown, and this question would come back later.

   Your answerA.Use a General Purpose SSD gp3 volume, because Multi-Attach became a General Purpose feature alongside provisioned throughput.Extends gp3's provisioning flexibility to a feature it does not have. The table marks Multi-Attach as not supported for both General Purpose types.The answerB.Use a Provisioned IOPS SSD volume, because Multi-Attach is supported on io1 and io2 and not on gp2 or gp3.The volume type table has a Multi-Attach row and it splits the SSD families.Your answerC.Use a Throughput Optimized HDD volume, because HDD volumes are shared by design across attached instances.Throughput Optimized HDD is marked as not supported for Multi-Attach, and HDD volumes are attached to one instance like any other volume.Your answerD.Use any volume type and attach it to one instance at a time, because Amazon EBS never supports concurrent attachment.Amazon EBS does support concurrent attachment through Multi-Attach on the Provisioned IOPS types; it is the General Purpose and HDD types that do not.

   Show the full explanation and doc quotes

   For gp3 and gp2 it reads "Amazon EBS Multi-attach: Not supported"; for io2 Block Express and io1 it reads "Supported". The HDD types, st1 and sc1, are also marked not supported. Multi-Attach lets a single Provisioned IOPS volume be attached to several instances in the same Availability Zone, and the table records that io2 Block Express additionally supports NVMe reservations, which is what a cluster-aware file system uses to coordinate access. The application's own software remains responsible for write coordination.

   Written from docs.aws.amazon.com/ebs/latest/userguide/ebs-volume-types.html

3. An architect sizes a gp3 volume for an analytics node that needs 2,000 MiB/s of throughput and asks what the volume must be provisioned with to reach it. Which solution meets these requirements?

   Right. In the game that’s Renown for your Design High-Performing Architectures building.

   Not that one: D is right. In the game your Design High-Performing Architectures building would lose a little Renown, and this question would come back later.

   Your answerA.A volume of at least 2,000 GiB, because gp3 throughput is provisioned at one MiB/s per GiB of volume size.Invents a size-to-throughput ratio. gp3 performance is independent of volume size.Your answerB.At least 80,000 provisioned IOPS and a volume of at least 160 GiB, because maximum throughput requires maximum IOPS.80,000 IOPS is available on volumes of 160 GiB or more, and throughput stops rising at 2,000 MiB/s long before that.Your answerC.A volume of at least 334 GiB, because that is the size at which General Purpose SSD reaches its throughput maximum.Describes gp2, where volumes of 334 GiB and larger deliver 250 MiB/s.The answerD.At least 8,000 provisioned IOPS and a volume of at least 16 GiB, because throughput is capped at 0.25 MiB/s per IOPS.The arithmetic is given directly. "You can provision additional throughput (up to a maximum of 2,000 MiB/s) for an additional cost at a ratio of 0.25 MiB/s per provisioned IOPS. Maximum throughput can be provisioned at 8,000 IOPS or higher and 16 GiB or larger (8,000 IOPS × 0.25 MiB/s per IOPS = 2,000 MiB/s)." So 8,000 IOPS is the floor that unlocks the 2,000 MiB/s ceiling, and the volume needs to be only 16 GiB for it - the size constraint bites on IOPS instead, where "maximum IOPS can be provisioned for volumes 160 GiB or larger."

   Written from docs.aws.amazon.com/ebs/latest/userguide/general-purpose.html