Testing Your Network

by Seth Noble |  Blog |  Jan 22, 2018

When I talk about high performance, I use terms like 100% utilization, filling the data path, maximum speed, and full capacity.  But just how fast can any given network path actually go?  Knowing a path's maximum speed is crucial to making decisions about how to improve real-world workflows.

The first place to look when determining your network's potential is the local WAN link.  Your ISP is being paid to provide a link at some number of megabits per second (mbps).  Make sure you know what that number is, then consider all the ways it might not mean what you think.

• Overhead — Headers for the various protocols consume a significant number of those bits: between 6% and 10%.  A link with a 100 mbps signal rate will typically deliver at most around 93 mbps of actual data transfer.  Adding protocol layers like VPNs, IPsec, MPLS, or PPPeE will lower it further.
• Asymmetry — Most consumer and many commercial links are much faster at downloading than uploading, typically about 5 times faster.  So if you've purchased a 100 mbps link from your ISP, don't be surprised if you only get 20 mbps uploading.
• Bursting — Most consumer and commercial, and even some enterprise links, do not have a fixed speed.  They may be faster for the first few seconds of a data transfer, then slow down substantially.  This can make testing tricky as short transfers may appear much faster than the link can sustain.
• LAN Speed — 10 gigabit per second WAN from the telecom, 1 gigabit per second switches on the LAN.  Yes, this happens a lot.  Even if your LAN should be faster than the uplink, local congestion and chokepoints can still slow down your data before it even leaves the building.
• The Other Side — All the factors we are talking about here affect the other end too.  Your download can only go as fast as their upload and vice-versa.

With all of that in mind, the easiest way to get a feel for whether or not your network is giving you all it can is to do some data transfers to and from nearby servers.  Sites like speedtest.net and fast.com can give you a rough idea of your network's capabilities.  But here again are some caveats:

• Legacy Protocols — Websites and conventional file transfer servers use TCP, which means that even if your local link is clear, the speed they measure is still going to vary a lot depending on distance to the server and mid-network congestion.  Do multiple tests at different times of day to get a feel for the variation.
• Throttling — The reason Netflix created fast.com is because not all ISPs treat all traffic the same, and some were caught giving preferential treatment to speedtest.net.  The same thing can happen with your own firewalls: giving preference to some traffic over others.
• Compression & Caching — Though rare on consumer broadband, satellite and some enterprise networks may have hidden compression and proxy devices.  These will change the speed you see depending on the content and protocols you use for testing.

When testing your network's capabilities, make sure to use as wide a variety of services, protocols, and test data as you can.  Just remember that the goal here is to test your network, not your storage (that is a whole different problem), so use the fastest, most directly attached storage you can (RAM and SSD being the best).

We provide a couple of test servers of our own, where you can download ExpeDat client software and then see how MTP/IP does on your network.  california.mtpip.net (1 gigabit per second) and florida.mtpip.net (100 megabits per second) may not be local to you, and their bandwidth is shared with other users, but with MTP/IP powering each transfer, they may give you a useful point of comparison to other test methods.  Installing your own ExpeDat trial server is also a quick way to establish a performance baseline.

After all these factors are taken into account, you can know the full capacity of your network.  This is what we mean when we say 100% utilization of available bandwidth.  (If that isn't close to what your ISP is promising, give them a call.)

Now you can compare that number to your real-world file transfer workflow.  Since you are visiting our site, it is a good bet that the speed there is significantly less than the 100% number we just figured out.  Latency, congestion, and the general inefficiency of TCP and legacy application protocols like FTP make the difference.  Removing those factors is the heart of what we do at Data Expedition, Inc.  MTP/IP cannot move data faster than the capacity of the physical network, but it can get you to that 100%.