What's the real bandwidth I should expect for a 64 GB RAM droplet?

June 24, 2017 903 views
Networking

What's the real egress and ingress bandwidth to and from the Internet I should expect for a 64 GB RAM droplet?

2 Answers

@yosucadilla

All Droplet's are connected to a 1Gbps port, though it's not a dedicated port as would be the case on a dedicated server where you're on your own :-).

Speed Tests for Data Centers

http://speedtest-nyc1.digitalocean.com/

Of course, the above are more so local tests (with a few smaller single file tests).

To test the connection from the outside within the Droplet, you'd probably want to test pulling down a file to the Droplet.

When pulling down the test 100 MB file on a fresh 64GB Droplet, it pulled down too quick to really do an accurate measure. It ran 107 MB/s, so the download took less than a second.

Pulling down a random 1 GB file from various locations resulted in the same overall average speed of 105-107 MB/s -- so between 9.5 and 9.9s.

...

The 64GB Droplet comes with 9TB of bandwidth. From my own experience, speeds are pretty consistent in and outbound

...

That said, depending on what the Droplet is being used for, a number of factors can affect speeds accordingly. Higher levels of traffic to the server could impact speeds as it would on any server and network connection, depending on what visitors are doing.

Is there any specific use case you have in mind?

  • Thank you jtittle,

    I just tested using the tool at the link you gave me, results for a 100MB file where:

    NYC1 DOWN: 8.97 Mb/s UP: 10.73 Mb/s
    NYC3 DOWN: 7.76 Mb/s UP: 7.49 Mb/s
    FRA1 DOWN: 7.77 Mb/s UP: 10.28 Mb/s
    LON1 DOWN: 13.30 Mb/s UP: 30.85 Mb/s

    You said: "Pulling down a random 1 GB file from various locations resulted in the same overall average speed of 105-107 MB/s -- so between 9.5 and 9.9s."

    107 MB/s is about 856 Mbps which is consistent with the advertised 1Gbit connections per physical machine.

    Why is there such a huge difference among your results and mine?
    Could it be that the connection between the physical node and the internet router maxes at 1 Gb so a small or medium instance only gets its fair share while the biggest servers would get access to the full interface?

    Just as you stated before: "All Droplet's are connected to a 1Gbps port, though it's not a dedicated port as would be the case on a dedicated server where you're on your own :-)."

    I need speeds over 70Mbps so these variations are really important for me.
    It's for a video router.

    • @yosucadilla

      Are you trying to download the test file locally? If so, you're ISP will play a role as you won't be able to go beyond their provided speeds, regardless of whether there's a 1 Gbps port or a 10 Gbps port on the physical node you're downloading from.

      For example, Charter Communications caps my speeds at around 60 Mbps down though from time to time, I can surpass that to the tune of squeezing 2-5 Mbps extra (62-65 Mbps). If I download one of the test files, I can max my connection download it without any issues.

      The average I noted in my previous response is direct download to the server, i.e.:

      wget http://speedtest-nyc1.digitalocean.com/100mb.test
      

      That should result in the same or close speeds as what I noted. Unless you have a 1 Gbit connection at your home or business, or even a 10 Gbit connection, chances are you won't see the same speeds as you would when downloading to the server directly as your ISP will always cap you.

      Server to server, in most all cases, is going to be the fastest. Anything else will depend on the client side and the client/visitors ISP or internet speed.

      ..

      As far as smaller Droplets, since there are more customers per server when it comes to the smaller sizes, their activity and usage also plays a role. The port is shared, so it's technically possible that someone could flood the host node uplink, though we do monitor such activity and if there is an issue, we take the appropriate measures to resolve it so that it doesn't affect others.

Thank you Jonathan, you were totally right.
I checked from the command line (from within the server) and speeds where something like 700Mb/s ingress and 270 Mb/s egress on a 2GB droplet.

Will these numbers increase with larger droplets in your experience?

  • @yosucadilla Remember to click the reply or use the @ to notify people :-)
    @jtittle There's another question for you, since I've never had that size droplet

  • @yosucadilla

    Currently, port speeds are set across the board, so you wouldn't be able to go beyond what is physically capped by the port whether it's a 512MB Droplet or a 64GB Droplet. Actual speeds vary with the number of Droplets to a hypervisor, though in all cases, networking in and out is monitored, so if there's an issue and something doesn't seem right, you're always more than welcome to submit a ticket and we can definitely check to see if there are any issues :-).

    Based on my experience, speeds are very consistent on all sizes, though they also depend on workload across the board. Since ports are shared and not physically dedicated, there's not a base speed guarantee. For a more guaranteed speed, you'd honestly need a dedicated port, though that's not something we're currently able to offer on Droplets and more so something that you'd see on a dedicated server where you're data is all that exists :-).

    That said, there's always options too -- if workload is too much for a given Droplet and you're seeing reduced speeds as a result, offloading via Load Balancing is an option -- in which case you'd be distributing load and network requests across multiple servers instead of just one.

    The setup is a bit more complex than a single Droplet, though more beneficial if you're truly concerned about saturation.

Have another answer? Share your knowledge.