AWS EBS Performance

Here is a very good presentation on Amazon EBS performance from reInvent in 2013:
http://www.slideshare.net/AmazonWebServices/ebs-webinarfinal

HPC for high IO

Enabled by Cluster family of EC2 instances, which includes:

• Cluster Compute instances: cc1.4xlarge,cc2.8xlarge
• Cluster GPU instances : cg1.4xlarge
• New c3.large, c3.xlarge, c3.2xlarge, c3.4xlarge and c3.8xlarge

Attributes of these instances:

• 10 Gbps Ethernet
• Latency and bandwidth about 10X better than “EC2 Classic” 
• Uses jumbo frames between cluster instances
• New EC2 API: Placement Groups. A placement group is a logical cluster 

Because HPC instance types have 10 Gbps networks, using 20 volumes attached to an HPC instance you can achieve in the neighborhood of 100 IOPS using 4K PIOPS volumes.

http://www.awsps.com.s3.amazonaws.com/training/computefest/AWS-HPC-BigData-Computefest.pdf

Data warehousing load times

Some times when people are moving to AWS for their databases they are looking for load times in the range of 2 TB per hour (this is 650+ MB/second).  

1. Green Plum : http://www.emc.com/collateral/hardware/data-sheet/h7419-greenplum-dca-ds.pdf : 10 TB to 60 TB per hour claimed in this document.

2. Exadata : http://www.oracle.com/ocom/groups/public/@otn/documents/webcontent/173705.pdf  Data Load Rate: Up to 12 TB/hour

Numbers listed in this book:  http://books.google.com/books?id=Whx9nldv8UEC&pg=PA102&lpg=PA102&dq=oracle+database+loads+terabytes+per+hour&source=bl&ots=VEXelUIhd2&sig=9LW4T0PgottDF2kpgvUobd2FonQ&hl=en&sa=X&ei=PHjyUcPKMuKbygHuyIDgBg&ved=0CHkQ6AEwBw#v=onepage&q=oracle%20database%20loads%20terabytes%20per%20hour&f=false

Of course,  Greenplum and Exadata are significant monetary investments.

Paraccel claims to achieve 2 TB an hour: http://www.paraccel.com/resources/Analyst-Reports/TDWI_Checklist-Report_Analytic-Databases-for-Big-Data.pdf

Redshift block size

Typical database block sizes range from 2 KB to 32 KB. Amazon Redshift uses a block size of 1 MB, which is more efficient and further reduces the number of I/O requests needed to perform any database loading or other operations that are part of query execution.

Oracle Database on ephemeral drives

Using EC2 ephemeral storage (either disk or SSD) is a way to achieve higher IO throughput.
You could use the design pattern Redshift uses (these use the HS1.* instances which have similar storage characteristics to the hi1.4xlarge instances) - "the first line of defense consists of two replicated copies of your data, spread out over up to 24 drives on different nodes within your data warehouse cluster".  This includes:

  1.  All data written to a node in your cluster is automatically replicated to other nodes within the cluster
  2.  All data is continuously backed up to Amazon S3

Oracle on SSD as it is recommendation to get highest level of IO when running Oracle on EC2.

Oracle Database on AWS - Striping disk

Achieving higher disk I/O when running Oracle on AWS on EC2 you can use Oracle Database ASM.  More information can be found here:

https://blogs.oracle.com/simonthorpe/entry/configuring_oracle_asm_disks_i