2023-01-21

spwnn - sweet spot

I was so excited that I could run a fast Graviton 3 64-core system for $2.30 / hour at AWS that I failed to consider where the sweet spot is for my app.


I'm pretty sure you can only get these things where the number of cores is a power of 2.  (Oops - not true - the sizes are 1, 2, 4, 8, 16, then increments of 16, so 32, 48, 64.)

From the graph, 16 cores is pretty fast,  16 cores is only $0.58 an hour!  It's up to me if I can stand to wait 17 seconds (16 cores) or 9.5 seconds (32 cores but twice the cost).  I dunno.  I guess I would try out 16 first and see how patient I am.

BTW, here's where to go to compare AWS instance types:  https://instances.vantage.sh/?filter=r6i

[Minor update: The largest Graviton 3 instance size as i write this and the one used above is the c7g.16xlarge.  I just tried out the 8xlarge version and 9 seconds to validate the dictionary is fine by me!  I'm about to try out c7g.4xlarge, and ..., that took 18 seconds, which felt okay.  I think if I was doing a lot of development the 4xlarge would be the right thing, because I would be spending 90% of my time in VIM, which barely uses one core.  And back to my point about these monster machines being available at hobbyist rates:  the 4xlarge is $0.60 an hour.] 


No comments:

Post a Comment