A bit confused on pricing so 1GB RAM with 1 Core (physical or virtual?) would be $50/month ?
andre1sk
sztosz
What does one CPU mean? How is it clocked? What’s its IPC? Basically, how many CPU instructions per second do I get?
1
jesse
@andre1sk @sztosz Unfortunately cgroups, and therefore docker, and therefore Kubernetes, and therefore Gigalixir makes understanding CPU shares a little confusing. The gist of it is, if you have N containers on a Kubernetes node and they are given weights [w(0)..w(N)], then the proportion of CPU container i gets is w(i)/sum(w). On Gigalixir, the most common machine type we use has 4 CPUS and 16GB of memory. I’ve tried to set it up so that when you buy 1GB of memory, you get 1/4 CPU to max utilize the machine. In practice, the machine is never “full” so you’ll get more than 1/4 CPU. In fact, Kubernetes ensures that the machine is never more than 80% utilized.
Basically, you get a proportion of CPU relative to what else is on the machine, but it will never be less than 1/4 CPU when you buy 1GB.
See the CPU shares section here for more: Understanding Linux Container Scheduling — Squarespace Engineering Blog
4
jesse
@sztosz The details of what 1 CPU is can be confusing as well. Google Compute Engine describes it here: CPU platforms | Compute Engine | Google Cloud Documentation
Basically, it is one hyperthread on one of the following, but you aren’t guaranteed a particular one because Gigalixir is run multi-zone.
- 2.6 GHz Intel Xeon E5 (Sandy Bridge)
- 2.5 GHz Intel Xeon E5 v2 (Ivy Bridge)
- 2.3 GHz Intel Xeon E5 v3 (Haswell)
- 2.2 GHz Intel Xeon E5 v4 (Broadwell)
- 2.0 GHz Intel Xeon (Skylake)
2
andre1sk
So it’s basically something like n1-standard-16 instance divided into 16 chunks (if we are using 1GB increments) ?
so 1GB would be 1 hyperThread ?
1
jesse
I think it’s actually a n1-standard-4 instance and I think it’s 1/4 of a hyperthread. GCE says, “For the n1 series of machine types, a virtual CPU is implemented as a single hardware hyper-thread on a 2.6 GHz Intel Xeon E5 (Sandy Bridge), 2.5 GHz Intel Xeon E5 v2 (Ivy Bridge), 2.3 GHz Intel Xeon E5 v3 (Haswell), 2.2 GHz Intel Xeon E5 v4 (Broadwell), or 2.0 GHz Intel Skylake (Skylake).”
The n1-standard-4 machine has 4 virtual CPUs so 4 hyperthreads on the physical hardware. Divided in 16 chunks makes 1/4 a hyperthread per chunk.
2
kordeviant
AWS is way too expensive, yet you people using google cloud infra should overprice your services.
I do not get it when you know dev ops so damn good, why not set up your data centers and break the Heroku prices.
this is the only way you would get extreme traction and rise up.
1
brightball
It’s the peripheral tools that suck people in. Managed and monitored instances of various databases, redis, memcached, bigquery, Aurora, etc in the same datacenter.
I’d love to see Gigalixir on Digital Ocean personally though.
1
brightball
When did Gigalixir add AWS? I just glanced at it again now that the conference is open and saw that there were two regions now.
That’s pretty great news.
1
zinclozenge
IBM kubernetes offering is apparently pretty good according to someone I know (and trust their opinion) so it might be a future target for gigalixir. Even better, they supposedly prorate so while you might provision 3x 4core16gb machines which would cost ~200+/mo if you’re only using ~25% cpu and memory then it might only cost ~$70-80.
0