As a note, if you find errors with the math below, please let me know.
This cost comparison only compares cost per I/O operation. It does not discuss storage or transit costs.
- Intel 320 series SSD, 300GB specs: http://www.intel.com/content/www/us/en/solid-state-drives/ssd-320-specification.html
At 4K, IOPS rates are advertised as:
- read: 39500/sec
- write: 23000/sec
Cost of an Intel 320 300GB drive: About $500
Comparing the same IOPS advertised by Intel 320 300GB, but with DynamoDB pricing - Pricing specifics are here: http://aws.amazon.com/dynamodb/#pricing
-
Read units: 39500 * 4 = 158000
-
Read cost: (158000 / 50) * 0.01 (per hour) = $31.60/hour
-
Monthly read cost: $31.60/hour * 24 hours * 30 days = $22752.00/month (every 30 days)
-
Write units: 23000 * 4 = 92000
-
Write cost: (92000 / 10) * 0.01 (per hour) = $92/hour
-
Monthly write cost: $92/hour * 24 hours * 30 days = $66240/month (every 30 days)
Amortizing the cost of hardware over 3 years is fairly common, so I will use that as a frame.
This puts the cost of the Intel SSD at $500 per 3 years.
The same IOPS performance on DynamoDB over 3 years (forgetting leap years) will cost:
- dynamodb total: $3,248,208 per 3 years
- intel total; $500 per 3 years.
separating reads and writes:
- dynamodb read cost: $31.60 * 24 * 365 * 3 = $830448
- dynamodb write cost: $92 * 24 * 365 * 3 = $2417760
Having started to ramp up close to $200 per month on dynamoDB, I've asked (and waiting for) Amazon to comment; meantime now rethinking as costs about to escalate. I invested in a migration and transformation from mysql to nosql but looking to host back on premise. AWS is expensive, I don't think this move is as stupid as it sounds - one rich corporate near me has done their research and decided on premise works better for them. I think the next few years, with graphene processors and the investment made by Apple, Google and Amazon in investment in open source technologies reduces the COO.