We're currently hosting a few low traffic web sites internally with a 1 Mbps upstream connection.

Is it a no brainer to go with Linode even if it adds a little latency (being located 2200 km away)?

Easy enough to figure out what the ballpark latency/throughput numbers will be for the datacenters.

http://www.linode.com/speedtest/

Seems to vary, but downloading the 100MB test file, I'm getting anywhere from 200-345 kB/s. 1 Mb/s = 128 kB/s, so Linode is able to push data about 2-3 times faster than my current connection. Maybe more since I probably don't get a full 128 kB/s upstream?

Depends on precisely what you're measuring. Communication links (similar to disks) are typically not rated in binary, so a 1Mbps upstream link is 1,000,000 bps. Then a TCP stream is only going to about 97% efficient for data transfer assuming a full 1500 byte MTU (a bit less if you only get 512 over a WAN connection).

So depending on what your measurement tool is showing (a typical download measurement is probably in term of data payload bytes, and with "K" = 1024), you should peak around 118KB/s (or 118KiB/s using the new fangled "kibi" prefix).

By default, a Linode will be able to generate 50Mbps in outbound bandwidth. Now depending on the pipe for a single user (bandwidth x latency) they may not see all of it, such as in your test, but multiple users can use up to that point depending on the load on your server, on average users might even see more improvement than you measure with a single test stream.

Note that the download test files are on hosts that aren't limited (I can get test downloads to a Linode in the same data center at 40-60MB/s). You can request to bump an individual Linode limit higher but have to show the ability to consume it.

At your peak rates in the ~350KB/s range are you saturating your download link? You should be able to get much better rates for the download files unless your distance is so far that you need larger than default TCP window sizes to keep the network pipe full. But for example, I can get 3+MB/s from a NY client to the Fremont data center which is a pretty long haul for latency.

-- David

Thanks for the info! My downstream is supposed to be 15Mbps. When I download multiple test files from different locations, the transfer rate drops below 100 kB/s in some cases. Is it safe to assume I'm saturating my download link?

I had someone download a test file from our old web server. He reported 64 kB/s. So 64 kB/s upstream, and about 400 kB/s +/- downstream. That's sad!

It would appear that Linode is the way to go! I wish I would've known about Linode a long time ago.

Well, a peak of ~400kB/s certainly isn't anywhere near 15Mbps (it's more like just over 3), but to be honest depending on the connection type and provider, and even time of day, what they claim you can get and what you can actually get can vary wildly. Though that seems far enough off that I'd consider it worth a problem report. For the Linode test downloads, not sure of your latency to the data center for your tests (my NY-Fremont test was about 95ms) but if that was very large it could have hurt the transfer a little but I don't think that much.

You could do some spot checks with speedtest.net to gauge the link itself - it's not a perfect test (and the Java application itself becomes a bottleneck on non-Windows platforms around 20-30Mbps), but it's a reasonable way to get a baseline and it should be reasonably accurate for a 15Mbps downstream link.

It's a pretty safe bet though that if you were to host the content on a Linode, that your users will be getting far more than 64kB/s downloads.

-- David