Benchmarking Vultr High Frequency Compute Instances
Hi everyone,
As you know, I’m working on Gladys Assistant, an open-source home automation software.
To control their home when they are away, Gladys users use the Gladys Gateway, an online proxy service I run, which allows them to access their local instance from anywhere in the world, all that end-to-end encrypted to respect their privacy.
Current hosting
Currently, the Gladys Gateway is hosted on 2 DigitalOcean droplets in Frankfurt datacenter.
- First droplet is running the PostgreSQL database.
- Second droplet is running 4 Docker containers:
- Redis
- Gladys Gateway server
- Nginx-proxy for SSL
- Let’s Encrypt companion to renew SSL certificate automatically
The problem
As we have more and more users using the Gladys Gateway + more and more users having cameras in Gladys, the usage is growing and our infrastructure needs to evolve.
Requirements
The server is mainly a websocket proxy, so what we want is:
- Located in France: For minimul latency, close to my users. I’ll had more locations if I have more users in other countries in the future, but that’s not the case right now. Frankfurt was already close, but we want closer.
- Lots of bandwidth: As we have camera image going through the gateway, we need > 1Gps port.
- Fast CPU: To handle lots of websockets messages in a short amount of time
- More RAMs: Websockets sessions are stored in Redis, in RAM for performance.
I looked at many providers :
- DigitalOcean: Their prices are great, performances are reliable, and the UI is just amazing. But they are not available in France (yet?) sadly.
- Scaleway : General purpose instances looked promising, but were expensive (Starting at 42$/month + VAT) + a quick twitter search showed me that Scaleway had reliability issue, not well handled… They don’t seem to care about uptime, which is very important for critical application like this. I want guarantees on SLA + automatic refund with a multiplier when the server is down.
- OVH: I read only bad feedbacks about them. Reliability & support is not their thing…
- Vultr: A great DigitalOcean-like provider, with a datacenter in France 😍 I saw that they launched a High Frequency compute offer recently, and this is what I’m going to benchmark here!
Vultr High Frequency Compute Benchmark
This is what they offer on their website (See Vultr High Compute Instances):
Exactly what we need for a high-usage websocket server!
The server
The server is a 2 vCPU + 4 Gb RAM Vultr High Compute Instance.
It started in 30 seconds, and I destroyed the server in 1 click after the tests. Running this server during tests cost me $0.04.
A few generic informations about the server :
Network speed
A little speedtest.net:
- Download: 3.5 Gb/s
- Upload: 4.3 Gb/s
Not bad !! 🚀
Ping
A ping from my place near Paris took 5ms on average:
In comparaison, a DigitalOcean instance in Franfkurt gives me a 20ms ping:
This is not that different if it was just a web server, but for real-time home automation, this is really better ! 🙂
Disk write performance
A simple dd of a 1 Gb file gives us some pretty decent performance :
Disk read performance
Reading the same file we just wrote gives us a 1.6 GB/s speed. Not bad !
CPU
I didn’t run any CPU tests, because there are plenty of them online (On VPSBenchmarks for example). Feedbacks on the High Frequency Compute are really great.
To Conclude
I already use Vultr to host Gladys Assistant community (with an amazing reliability over years), and I have to say that once again, they proved that they can provide performant cloud instances.
I would love your feedback on this, but I think I’ll migrate Gladys Gateway servers to Vultr soon.
I want to improve latency and speed for my users, and I’m convinced that this change will help us in this way !
If you loved this story, you can subscribe to my newsletter here.