5

I have a google compute engine instance in us-central1-b running Centos 7 with 1vCPU and 3.75GB RAM.

According to the section "Egress throughput caps" on https://cloud.google.com/compute/docs/networks-and-firewalls my internet speed should be something somewhat close to 2Gb ("Each core is subject to a 2 Gbits/second (Gbps) cap for peak performance.").

I have installed speedtest-cli (pip install speedtest-cli) and performed many tests at different times of day. My max download has been 500Mb and max upload has been 72Mb.

I don't really expect to get 2Gb on this little VM, but I would expect it to be much closer to 1Gbps and a more synchronous connection.

Have I missed some documentation somewhere explaining the WAN speeds better? Am I not understanding the referenced documentation correctly? Is my system somehow misconfigured causing these really slow asynchronous speeds?

Thanks for any feedback.

Also, there is no other traffic on this system; it's just a test box.

  • 1
    Have you tried against any other benchmarks? Maybe the bottleneck is the other end, not your end in google? – leinad13 Dec 22 '16 at 15:37
  • 5
    Those numbers are relevant only for the Google network itself. Those numbers say nothing about any other network, which will have some bearing on your speed tests. Where is the remote end of your test? What is the network quality,latency, etc. between your Google instance and the remote end? Etc., etc. My suggestion would be to spin up a second Google instance of the same type and size and perform a speed test between the two instances. That will tell you if Google is living up to their claim. – joeqwerty Dec 22 '16 at 15:44
  • Here are the results from several speed tests to many different locations as well as the details about the location. – Jeff Dec 22 '16 at 15:51
  • 1
    Right, but again those speed test results are dependent upon the quality of external networks beyond the control of Google. The WAN egress caps are relevant only to traffic within the Google network. That's why you need to perform a test between two instances within the Google network. – joeqwerty Dec 22 '16 at 16:01
  • 1
    The article states it as being egress traffic from a virtual machine, not from the Google network. – joeqwerty Dec 22 '16 at 16:35
5

As per the documentation, "All caps are meant as maximum possible performance between two VM instances in the same zone, both using internal IP addresses." I verified running two instances running Centos 7 with 1vCPU and 3.75GB RAM in same zone using the "iperf" [1][2] (sudo yum install iperf) and received 1.64 Gbps.

On the GCE VM machine (with iperf install) in the same zone, run the 
command:
iperf -s 

On egress test GCE VM machine (with iperf install) run the command 
iperf -c xx.xx.xx.xx     (where xx.xx.xx.xx is internal IP of the other VM- 
instance)

output:
[ID] Interval       Transfer     Bandwidth
[3]  0.0-10.0 sec  1.91 GBytes  1.64 Gbits/sec

I verified running instance with 4 vCPUs as well and received 7.67 Gbits/sec as bandwidth which is very close to 8 Gbps as defined in the documentation. I would suggest you to verify the same.

Moreover, as Google does not have any control over ISP network and other factors involved (eg: routing etc.) over internet traffic, Google cannot guarantee the egress throughput for the egress network traffic from a virtual machine over the ISP network.

[1] https://iperf.fr/

[2] https://www.linode.com/docs/networking/diagnostics/diagnosing-network-speed-with-iperf

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.