Ask questions about dedicated servers here and we and other users will do our best to answer them. Please also refer to the self-help section for tutorials and answers to the most commonly asked questions.
Edge100x wrote:I didn't read this closely enough the first time and thought that you were asking for people to test their NFO service's speeds with a separately-hosted tester. I believe now that you are saying you are hosting actual speed tests on our network.
This may not be something that we can effectively support, for a few reasons:
- Limited bandwidth. If you run a testing service on a VDS, you will be limited to the 1 Gbps network adapter on the machine, which means that if two people are using your tool at the same time, the answer will be off. This could cause the test to misrepresent the client's real speed and/or the speed of our network.
- Performance degradation for other customers. Speed tests are often designed to consume as much bandwidth as they possibly can and get around mechanisms that back off transfer speeds (specifically, some use UDP floods so that TCP congestion-avoidance algorithms don't come into play; others try to compensate for TCP's exponential backoff system by using multiple simultaneous TCP streams and trying to estimate what the real ceiling is). Unfortunately, the nature of this means that other VDS customers on the same physical machine could experience degraded performance while a test is in progress, even if only one person is testing. Normal applications don't cause this same problem. I don't know how your specific test is designed, so I don't know how much this comes into play.
- Support. If hosted speed tests mention us, users may have questions and concerns about their results, but because we don't run them, we won't have answers for them. Some users will see lower speeds than they expect, for instance, and want to know why. This may not be a problem if there aren't many people coming in and we're fixing real routing problems for them.
(Speed tests in general can be a bit misleading because they test the entire path to/from the client and server, which could bottleneck within the customer's ISP, specific peering/transit points of his ISP, or peering between specific NSPs. Many clients don't realize this, and a common misunderstanding is that the single output number represents their overall connection quality or the host's overall connection quality. But, you know that )
If you wish to run test sites, it would be best to have them on 10 Gbps connections and their own hardware. Our bandwidth (and Internap) will be about as good as it gets for trying to eliminate other bottlenecks, so that component is relatively solid.
I guess we'll have to see how things go and see if these end up being real concerns.
Here's what I was able to pull from my VDS in Chicago;
speedtest-cli
Retrieving speedtest.net configuration...
Retrieving speedtest.net server list...
Testing from Nuclearfallout Enterprises (192.xxx.xx.xxx)...
Selecting best server based on latency...
Hosted by Host-Engine.com (Chicago, IL) [3.75 km]: 6.867 ms
Testing download speed........................................
Download: 916.36 Mbit/s
Testing upload speed..................................................
Upload: 445.56 Mbit/s
I think NFO should host a speedtest service here that only connections from the NFO network could access for quality and accurate results and services like http://www.speedtest.net/ are unreliable as most the servers they use are always loaded and sometimes cant handle more then a certain amount of bandwidth.
Running a Speedtest server is something we've talked about doing previously. The biggest issue is that they are big bandwidth suckers and can cause temporary performance impacts for other customers on the box. Not to mention you'd need several 10Gbps NICs to ensure multiple people could run at the same time without seeing a poor results, and those are pretty expensive.
@Kraze^NFo> Juski has a very valid point
@Juski> Got my new signature, thanks!
@Kraze^NFo> Out of context!
@Juski> Doesn't matter!
@Juski> You said I had a valid point! You can't take it back now! It's out there!
Yes, it possibly wasn't the best idea to quote that specific post, I was more or less trying to point out the shared gigabit port and how other customers using the same physical adapter simultaneously may hinder the speed-test results, generally speaking we know these types of tests are not an accurate way in measuring the throughput of such speeds, being that the host of the test server itself may have bandwidth and or hardware limitations.