AWS EBS Performance

AWS EBS Performance Tips

  • EBS Performance depends on several factors including I/O characteristics,  instances and volumes configuration and can be improved using PIOPS, EBS-Optimized instances, Pre-Warming, and RAIDed configuration.

EBS-Optimized or 10 Gigabit Network Instances

  • An EBS-Optimized instance uses an optimized configuration stack and provides additional, dedicated capacity for EBS I/O.
  • Optimization provides the best performance for the EBS volumes by minimizing contention between EBS I/O and other traffic from an instance
  • EBS-Optimized instances deliver dedicated throughput to EBS depending on the instance type used.
  • Not all instance types support EBS-Optimization
  • Some Instance types enable EBS-Optimization by default, while it can be enabled for some.
  • EBS optimization enabled for an instance, that is not EBS-Optimized by default, an additional low, hourly fee for the dedicated capacity is charged
  • When attached to an EBS–optimized instance,
    • General Purpose (SSD) volumes are designed to deliver within 10% of their baseline and burst performance 99.9% of the time in a given year
    • Provisioned IOPS (SSD) volumes are designed to deliver within 10% of their provisioned performance 99.9 percent of the time in a given year.

EBS Volume Initialization – Pre-warming

  • Empty EBS volumes receive their maximum performance the moment that they are available and DO NOT require initialization (pre-warming).
  • EBS volumes needed a pre-warming, previously, before being used to get maximum performance to start with. Pre-warming of the volume was possible by writing to the entire volume with 0 for new volumes or reading the entire volume for volumes from snapshots.
  • Storage blocks on volumes that were restored from snapshots must be initialized (pulled down from S3 and written to the volume) before the block can be accessed.
  • This preliminary action takes time and can cause a significant increase in the latency of an I/O operation the first time each block is accessed.
  • To avoid this initial performance hit in a production environment, the following options can be used
    • Force the immediate initialization of the entire volume by using the dd or fio utilities to read from all of the blocks on a volume.
    • Enable fast snapshot restore – FSR on a snapshot to ensure that the EBS volumes created from it are fully-initialized at creation and instantly deliver all of their provisioned performance.

RAID Configuration

  • EBS volumes can be striped, if a single EBS volume does not meet the performance and more is required.
  • Striping volumes allows pushing tens of thousands of IOPS.
  • EBS volumes are already replicated across multiple servers in an AZ for availability and durability, so AWS generally recommend striping for performance rather than durability.
  • For greater I/O performance than can be achieved with a single volume, RAID 0 can stripe multiple volumes together; for on-instance redundancy, RAID 1 can mirror two volumes together.
  • RAID 0 allows I/O distribution across all volumes in a stripe, allowing straight gains with each addition.
  • RAID 1 can be used for durability to mirror volumes, but in this case, it requires more EC2 to EBS bandwidth as the data is written to multiple volumes simultaneously and should be used with EBS–optimization.
  • EBS volume data is replicated across multiple servers in an AZ to prevent the loss of data from the failure of any single component
  • AWS doesn’t recommend RAID 5 and 6 because the parity write operations of these modes consume the IOPS available for the volumes and can result in 20-30% fewer usable IOPS than RAID 0.
  • A 2-volume RAID 0 config can outperform a 4-volume RAID 6 that costs twice as much.

RAID Configuration

AWS Certification Exam Practice Questions

  • Questions are collected from Internet and the answers are marked as per my knowledge and understanding (which might differ with yours).
  • AWS services are updated everyday and both the answers and questions might be outdated soon, so research accordingly.
  • AWS exam questions are not updated to keep up the pace with AWS updates, so even if the underlying feature has changed the question might not be updated
  • Open to further feedback, discussion and correction.
  1. A user is trying to pre-warm a blank EBS volume attached to a Linux instance. Which of the below mentioned steps should be performed by the user?
    1. There is no need to pre-warm an EBS volume (with latest update no pre-warming is needed)
    2. Contact AWS support to pre-warm (This used to be the case before, but pre warming is not necessary now)
    3. Unmount the volume before pre-warming
    4. Format the device
  2. A user has created an EBS volume of 10 GB and attached it to a running instance. The user is trying to access EBS for first time. Which of the below mentioned options is the correct statement with respect to a first time EBS access?
    1. The volume will show a size of 8 GB
    2. The volume will show a loss of the IOPS performance the first time (the volume needed to be wiped cleaned before for new volumes, however pre warming is not needed any more)
    3. The volume will be blank
    4. If the EBS is mounted it will ask the user to create a file system
  3. You are running a database on an EC2 instance, with the data stored on Elastic Block Store (EBS) for persistence At times throughout the day, you are seeing large variance in the response times of the database queries Looking into the instance with the isolate command you see a lot of wait time on the disk volume that the database’s data is stored on. What two ways can you improve the performance of the database’s storage while maintaining the current persistence of the data? Choose 2 answers
    1. Move to an SSD backed instance
    2. Move the database to an EBS-Optimized Instance
    3. Use Provisioned IOPs EBS
    4. Use the ephemeral storage on an m2.4xLarge Instance Instead
  4. You have launched an EC2 instance with four (4) 500 GB EBS Provisioned IOPS volumes attached. The EC2 Instance is EBS-Optimized and supports 500 Mbps throughput between EC2 and EBS. The two EBS volumes are configured as a single RAID 0 device, and each Provisioned IOPS volume is provisioned with 4,000 IOPS (4000 16KB reads or writes) for a total of 16,000 random IOPS on the instance. The EC2 Instance initially delivers the expected 16,000 IOPS random read and write performance. Sometime later in order to increase the total random I/O performance of the instance, you add an additional two 500 GB EBS Provisioned IOPS volumes to the RAID. Each volume is provisioned to 4,000 IOPS like the original four for a total of 24,000 IOPS on the EC2 instance Monitoring shows that the EC2 instance CPU utilization increased from 50% to 70%, but the total random IOPS measured at the instance level does not increase at all. What is the problem and a valid solution?
    1. Larger storage volumes support higher Provisioned IOPS rates: increase the provisioned volume storage of each of the 6 EBS volumes to 1TB.
    2. EBS-Optimized throughput limits the total IOPS that can be utilized use an EBS-Optimized instance that provides larger throughput. (EC2 Instance types have limit on max throughput and would require larger instance types to provide 24000 IOPS)
    3. Small block sizes cause performance degradation, limiting the I’O throughput, configure the instance device driver and file system to use 64KB blocks to increase throughput.
    4. RAID 0 only scales linearly to about 4 devices, use RAID 0 with 4 EBS Provisioned IOPS volumes but increase each Provisioned IOPS EBS volume to 6.000 IOPS.
    5. The standard EBS instance root volume limits the total IOPS rate, change the instant root volume to also be a 500GB 4,000 Provisioned IOPS volume
  5. A user has deployed an application on an EBS backed EC2 instance. For a better performance of application, it requires dedicated EC2 to EBS traffic. How can the user achieve this?
    1. Launch the EC2 instance as EBS provisioned with PIOPS EBS
    2. Launch the EC2 instance as EBS enhanced with PIOPS EBS
    3. Launch the EC2 instance as EBS dedicated with PIOPS EBS
    4. Launch the EC2 instance as EBS optimized with PIOPS EBS

AWS EC2 Instance Types

EC2 Instance Types

  • EC2 Instance types determine the hardware of the host computer used for the instance.
  • EC2 Instance types offer different compute, memory & storage capabilities and are grouped in instance families based on these capabilities
  • EC2 provides each instance with a consistent and predictable amount of CPU capacity, regardless of its underlying hardware.
  • EC2 dedicates some resources of the host computer, such as CPU, memory, and instance storage, to a particular instance.
  • EC2 shares other resources of the host computer, such as the network and the disk subsystem, among instances. If each instance on a host computer tries to use as much of one of these shared resources as possible, each receives an equal share of that resource. However, when a resource is under-utilized, an instance can consume a higher share of that resource while it’s available.

NOTE – AWS keeps on releasing new instances types, please refer AWS documentation for the same.

EC2 Instance Types – Current Generation

EC2 Instance Types

EC2 Instance Types Selection criteria

  • Some Instance types support only the HVM virtualization type while others support both the PV and HVM virtualization types. AWS, however, recommends using HVM for taking advantage of the underlying hardware
  • All EC2 instance types are available in a VPC, however, a few are not available in an EC2-classic. AWS recommends using VPC to take advantage of enhanced networking, multiple IP addresses, finer security control etc.
  • Some instances support only EBS volumes, while others support both EBS and Instance store volumes. Some instances that support instance store volumes use solid-state drives (SSD) to deliver very high random I/O performance.
  • Some EC2 instance types can be launched as EBS optimized instances with a dedicated capacity for EBS I/O.
  • Some EC2 Instance types can be launched in placement group to optimize instances for High-Performance Computing (HPC)
  • Some instances support Enhanced Networking,  to get significantly higher packet per second (PPS) performance, lower network jitter, and lower latencies
  • Some Instances allow EBS volumes to be encrypted

EBS-Optimized

  • EBS-optimized instance uses an optimized configuration stack and provides additional, dedicated capacity for EBS I/O
  • EBS-optimized instances enable you to get consistently high performance for the EBS volumes by eliminating contention between EBS I/O and other network traffic from the instance
  • EBS-optimized instances deliver dedicated throughput between Amazon EC2 and EBS, with options between 500 and 60,000 Megabits per second (Mbps) depending on the instance type used.
  • When attached to an EBS–optimized instance, General Purpose (SSD) volumes are designed to deliver within 10 percent of their baseline and burst performance 99.9 percent of the time in a given year, and Provisioned IOPS (SSD) volumes are designed to deliver within 10 percent of their provisioned performance 99.9 percent of the time in a given year.
  • EBS optimization can be enabled for an instance that is not EBS–optimized, by default

Placement Groups

Refer to My Blog Post @ EC2 Placement Groups

EC2 Instance Types Comparision

Screen Shot 2016-04-15 at 7.06.50 AM.png

T2 Instances (General Purpose)

  • T2 instances are designed to provide moderate baseline performance and the capability to burst to significantly higher performance as required
  • Mainly intended for workloads that don’t use the full CPU often or consistently, but occasionally need to burst.
  • T2 instances are well suited for
    • general-purpose workloads, such as web servers, developer environments, remote desktops, and small databases
  • Requirements
    • can be launched only with HVM AMI
    • can be launched into a  VPC only, and not supported on the EC2-Classic platform
    • are available as EBS-backed instances only
    • are available as On-Demand, Reserved instances, Dedicated Instances (T3 only), and Spot instances but do not allow spot instances
    • By default, 20 (soft limit) T2 instances can run simultaneously
    • cannot be launched as a Dedicated host
  • T2 Unlimited Instances
    • can sustain high CPU performance for as long as a workload needs it.
    • for most general-purpose workloads, it provides ample performance without any additional charges.
    • If the instance needs to run at higher CPU utilization for a prolonged period, it can also do so at a flat additional rate

CPU Credits

  • CPU Credits (Similar to I/O Credits in the case of the EBS general-purpose storage) provides the performance of a full CPU core for one minute
  • T2 instances provide a baseline level of CPU performance, while CPU governs the ability to burst above the baseline level
  • One CPU credit is equal to one vCPU running at 100% utilization for one minute. for e.g. can have One vCPU running at 100% for One min OR One vCPU running @ 50% for 2 mins OR Two vCPU running @ 25% for 2 mins
  • Each T2 instance receives a healthy initial credit balance for startup performance
  • Initial CPU credits do not expire, but they are used first when an instance uses CPU credits.
  • Each T2 instance then continuously (at a millisecond-level resolution) receives a set rate of CPU credits per hour, depending on instance size for e.g. t2.nano earns 3/hour while a t2.large earns 36/hour
  • Each T2 instance accumulates the CPU credit when it uses fewer CPU resources than its allowed baseline performance levels
  • Maximum earned credit balance for an instance is equal to the number of CPU credits received per hour times 24 hours for e.g. t2.nano can earn max 72 (24 * 3) credits
  • CPU credit balance is available for a period of 24 hours and it expires 24 hours after they were earned. Expired credits are removed from the balance before new ones are added
  • CPU credit ceases to persist between an instance stop-start. However, after the start, the instance receives the initial CPU credits again
  • When the credit balance is completely exhausted, the instance will perform at its baseline performance

C4 Instances (Compute Intensive)

  • C4 instances are ideal for compute-bound applications that benefit from high-performance processors
  • Well suited for
    • Batch processing workloads,
    • Media transcoding,
    • High-traffic web servers, massively multiplayer online (MMO) gaming servers, and ad serving engines,
    • High-traffic web servers, massively multiplayer online (MMO) gaming servers, and ad serving engines
  • Features
    • are EBS-optimized, by default
    • can be enabled for Enhanced Networking capabilities
    • can be clustered in a placement group
  • requirements
    • requires 64-bit HVM AMI
    • can be launched into a  VPC only, and not supported on the EC2-Classic platform

G2 Instances (Graphic Intensive)

  • GPU instances provide  high parallel processing capability
  • Well suited for
    • to accelerate many scientific, engineering, and rendering applications by leveraging the Compute Unified Device Architecture (CUDA) or OpenCL parallel computing frameworks
    • graphics applications, including game streaming, 3-D application streaming, and other graphics workloads
  • Requirements
    • requires HVM AMI
    • can’t access GPU unless NVIDIA drivers installed
  • Features
    • can be clustered in a placement group

I2 Instances (I/O Intensive)

  • I2 instances are optimized to deliver tens of thousands of low-latency, random I/O operations per second (IOPS) to applications.
  • Well suited for applications
    • NoSQL databases (for example, Cassandra and MongoDB)
    • Clustered databases
    • Online transaction processing (OLTP) systems
  • Features
    • Primary data storage is SSD-based instance storage.
    • can be enabled for Enhanced Networking capabilities
    • can be clustered in a placement group
    • can enable EBS–optimization to obtain additional, dedicated capacity for Amazon EBS I/O
  • Requirements
    • requires HVM AMI
  • HI1 is the equivalent previous generation instance
    • supports both PV and HVM AMIs

D2 Instances (Density Intensive)

  • D2 instances are designed for workloads with very high storage density and that require high sequential read and write access to very large data sets on local storage.
  • Well suited for applications
    • Massive parallel processing (MPP) data warehouse
    • MapReduce and Hadoop distributed computing
    • Log or data processing applications
  • Features
    • Primary data storage for D2 instances is HDD-based instance storage
    • are EBS-optimized, by default
    • can be enabled for Enhanced Networking capabilities
    • can be clustered in a placement group
  • requirements
    • requires 64-bit HVM AMI
  • HS1 is the equivalent previous generation instance
    • supports both EBS and Instance store backed AMIs
    • supports both PV and HVM AMIs

AWS Certification Exam Practice Questions

  • Questions are collected from Internet and the answers are marked as per my knowledge and understanding (which might differ with yours).
  • AWS services are updated everyday and both the answers and questions might be outdated soon, so research accordingly.
  • AWS exam questions are not updated to keep up the pace with AWS updates, so even if the underlying feature has changed the question might not be updated
  • Open to further feedback, discussion and correction.
  1. Which of the following instance types are available as Amazon EBS-backed only? Choose 2 answers
    1. General purpose T2
    2. General purpose M3
    3. Compute-optimized C4
    4. Compute-optimized C3
    5. Storage-optimized 12
  2. A t2.medium EC2 instance type must be launched with what type of Amazon Machine Image (AMI)?
    1. An Instance store Hardware Virtual Machine AMI
    2. An Instance store Paravirtual AMI
    3. An Amazon EBS-backed Hardware Virtual Machine AMI
    4. An Amazon EBS-backed Paravirtual AMI
  3. You have identified network throughput as a bottleneck on your m1.small EC2 instance when uploading data Into Amazon S3 In the same region. How do you remedy this situation? Add an additional ENI
    1. Change to a larger Instance
    2. Use DirectConnect between EC2 and S3
    3. Use EBS PIOPS on the local volume
  4. You are using an m1.small EC2 Instance with one 300 GB EBS volume to host a relational database. You determined that write throughput to the database needs to be increased. Which of the following approaches can help achieve this? Choose 2 answers
    1. Use an array of EBS volumes (Striping to increase throughput)
    2. Enable Multi-AZ mode.
    3. Place the instance in an Auto Scaling Groups
    4. Add an EBS volume and place into RAID 5 (RAID 5 is not recommended as it provides parity and EBS volumes are already replicated across multiple servers in an Availability Zone for availability and durability, so AWS recommends striping for performance rather than durability)
    5. Increase the size of the EC2 Instance.
    6. Put the database behind an Elastic Load Balancer.
  5. You are tasked with setting up a cluster of EC2 Instances for a NoSQL database. The database requires random read IO disk performance up to a 100,000 IOPS at 4KB block side per node. Which of the following EC2 instances will perform the best for this workload?
    1. A High-Memory Quadruple Extra Large (m2.4xlarge) with EBS-Optimized set to true and a PIOPs EBS volume
    2. A Cluster Compute Eight Extra Large (cc2.8xlarge) using instance storage
    3. High I/O Quadruple Extra Large (hi1.4xlarge) using instance storage
    4. A Cluster GPU Quadruple Extra Large (cg1.4xlarge) using four separate 4000 PIOPS EBS volumes in a RAID 0 configuration
  6. You are implementing a URL whitelisting system for a company that wants to restrict outbound HTTP’S connections to specific domains from their EC2-hosted applications you deploy a single EC2 instance running proxy software and configure It to accept traffic from all subnets and EC2 instances in the VPC. You configure the proxy to only pass through traffic to domains that you define in its whitelist configuration You have a nightly maintenance window or 10 minutes where ail instances fetch new software updates. Each update Is about 200MB In size and there are 500 instances In the VPC that routinely fetch updates After a few days you notice that some machines are failing to successfully download some, but not all of their updates within the maintenance window The download URLs used for these updates are correctly listed in the proxy’s whitelist configuration and you are able to access them manually using a web browser on the instances What might be happening? (Choose 2 answers) [PROFESSIONAL]
    1. You are running the proxy on an undersized EC2 instance type so network throughput is not sufficient for all instances to download their updates in time.
    2. You have not allocated enough storage to the EC2 instance running me proxy so the network buffer is filling up causing some requests to fall
    3. You are running the proxy in a public subnet but have not allocated enough EIPs to support the needed network throughput through the Internet Gateway (IGW)
    4. You are running the proxy on a affluently-sized EC2 instance in a private subnet and its network throughput is being throttled by a NAT running on an undersized EC2 instance
    5. The route table for the subnets containing the affected EC2 instances is not configured to direct network traffic for the software update locations to the proxy.
  7. You have been asked to design the storage layer for an application. The application requires disk performance of at least 100,000 IOPS in addition; the storage layer must be able to survive the loss of an individual disk, EC2 instance, or Availability Zone without any data loss. The volume you provide must have a capacity of at least 3TB. Which of the following designs will meet these objectives? [PROFESSIONAL]
    1. Instantiate an i2.8xlarge instance in us-east-1a. Create a RAID 0 volume using the four 800GB SSD ephemeral disks provided with the instance. Provision 3×1 TB EBS volumes attach them to the instance and configure them as a second RAID 0 volume. Configure synchronous, block-level replication from the ephemeral backed volume to the EBS-backed volume. (Same AZ will not survive the AZ loss)
    2. Instantiate an i2.8xlarge instance in us-east-1a. Create a RAID 0 volume using the four 800GB SSD ephemeral disks provided with the Instance Configure synchronous block-level replication to an identically configured Instance in us-east-1b.
    3. Instantiate a c3.8xlarge Instance in us-east-1. Provision an AWS Storage Gateway and configure it for 3 TB of storage and 100,000 IOPS. Attach the volume to the instance. (Need synchronous replication to prevent any data loss)
    4. Instantiate a c3.8xlarge instance in us-east-1 provision 4x1TB EBS volumes, attach them to the instance, and configure them as a single RAID 5 volume Ensure that EBS snapshots are performed every 15 minutes. (RAID 5 not recommended by AWS and Need synchronous replication to prevent any data loss)
    5. Instantiate a c3 8xlarge Instance in us-east-1 Provision 3x1TB EBS volumes attach them to the instance, and configure them as a single RAID 0 volume Ensure that EBS snapshots are performed every 15 minutes. (Need synchronous replication to prevent any data loss)

References