How much does Spoke cost to host?

[schuyler1d]

This is currently somewhat documented here (might be OUTDATED):

• https://opensource.moveon.org/estimated-cost

This info is subject to change when vendors change their pricing, so dates should be recorded for estimates.

General Performance guidelines for cost estimates (2021-05-11)

• The largest cost of any texting campaign will be per-message texting through the service vendor (e.g. Twilio). Costs are fluctuating at this moment due to carriers incresaing their fees. https://www.twilio.com/sms/pricing/us
• Server costs will generally be less than 20% of texting costs. For those without fixed technical expertise, adding technical expertise (i.e. hiring a developer and/or sysadmin) can significantly increase the cost of running and maintaining Spoke. This document will not account for staffing costs.
• Spoke's performance is generally limited by Database CPU and Redis (caching) memory size. All other costs will be at least one order of magnitude smaller. Spoke scales linearly (that's a good thing) with these resources -- so if X resources can handle Y contacts-per-day, then 2 X resources can handle 2 Y contacts-per-day.

Estimates for good 'napkin calculations'

• At the DB level for storage, assume ~3Kb/contact (including messages, etc)
• Redis should be resourced for ~1K/contact for the maximum contact count that will be sent messages across a day. (so e.g. a maximum day of 1MM contacts should have a redis instance of ~ 1Gb)
• A db of AWS db.r4.xlarge with 5000 IOPS provisioned can handle about 1MM messages/day (or ~ 500-700K/hour)

[schuyler1d]

Amazon Lambda Deployment

Pricing sheets (note that these vary by-region):

• Database pricing
  • Aurora pricing: https://aws.amazon.com/rds/aurora/pricing/
  • 8 ACU (min / max) should handle ~250k messages an hour, 16 ACU ~500k per-hour
  • "classic" RDS instance pricing: https://aws.amazon.com/rds/postgresql/pricing/
• Redis Elasticache pricing: https://aws.amazon.com/elasticache/pricing/

Choosing between Aurora serverless and 'classic' RDS has some subtle pros/cons depending on your org's (tech staff's) ability to ramp up/down manually depending on each day's planned scale of texting. Amazon advertises Aurora as having the ability to 'auto-scale' however, in-practice with the Spoke application if/when Aurora 'autoscales' during texting ramp up (your thundering herd of texters), Spoke is observed to 'lock up' right at the worst moment. So IF you choose Aurora, it's better to 'peg' it for your anticipated scale of contacts for the day -- before your texters 'arrive'. Most often, campaigns will know what scale of texting will occur for the week or day -- if your internal workflow can accomodate technical staff having the capacity and reliably informed by your campaign staff what scale they will be at, then this can be very cost-effective (pay for much less when not texting, and then only ramp up costs when actively texting at the appropriate scale).

To avoid all of that, it's often possible to find out what the maximum texting campaign will be "this year" -- and then resource an RDS instance for that. This is similar to phone companies and "Mother's day" -- i.e. you plan for maximum usage rather than average usage, to avoid failures.

While Aurora (serverless or not) charges slightly more per CPU-level than RDS, at scales of > 1 million messages-per-day, we recommend Aurora anyway since it has 'limitless' IOPS (IO operations per-second) which can become a bottleneck at higher scales.