Broadband … really?

How does one define broadband?  Is it all defined around throuput?  Latency?  Between which endpoints?  Using ADSL currently one can expect a local latency of around 30ms to anywhere within SA, and depending on whether you’ve got a SAIX (Fibre) or IS (both Fibre and Satellite) you can expect anything between 300ms and 800ms to international destinations (obviously depending on the exact geographical location and the type/distance of the peerings being used).

What kinds of speeds is “fast enough”?  What is the trade-off between throughput and latency?

Personally, I could do with >4Mbps links, but this isn’t essential for me.  Heck, I still use a 384 DSL line at home.  However, I obviously won’t be running a company off of it.  Nor am I a big leecher that tries to download a gazillion gigabytes every month.

Half the time, I find that people that complain about “slow” internet isn’t even saturating their link at all.  So why do they consider their links to be slow?  Simple, latency.  Let’s say you’re browsing to, (hope page, excluding images is 2909 bytes, that’s _two_ TCP segments).  At 4Mbps that’s a fraction of a second, then why does it feel like takes a second to load?  Because it actually does:

  1. SYN packet gets sent.
  2. ~300ms later the SYN/ACK arrives.
  3. ACK gets sent + request gets sent off.
  4. ~300ms later the response starts coming in.

So yes, that’s around 600ms (on a low-latency round trip) before we even START receiving the data from google back.  Please note that local caches and other factors obviously influences this.

Then, there is probably the single factor that I consider to be the most crucial piece of the puzzle at the moment:  packet loss!  I’ve recently started seeing round-trip packet loss of around 5 %!  Sounds low?  Well, it’s actually less than 5 % as the 5 % is round-trip, so it’s probably like 2.6 % per direction.  Either way, for VoIP we are trying to send 50 packets per direction per second, at 2 % that means we’re losing a fragment per second (0.02ms worth of audio).

As previously stated we can get away with relatively little bandwidth for VoIP, but there are a few things we do want:

  1. Lowish (<100ms) latency.
  2. Low jitter (The lower ping’s mdev value, the better, <5ms works well).
  3. NO PACKET LOSS.  We do NOT want to lose packets at all.

It is mostly for the latter two reasons that we HIGHLY recommend dedicated internet links for VoIP purposes in SA (No QoS on the network).  However, it seems recently the major providers started competing to see who can provide the highest latency (iBurst is mostly winning here at around 150ms+ latency for local traffic) and packet loss (it’s hard to say who’s leading here … I saw iBurst at 40% the other weekend, I’m getting more DSL lines at 5%+), and let’s not even go into jitter …

No, internet in South Africa at this point in time is a mixture of disappointment and utter frustration.  I’ve been with my hands in my hair the last while, and I’ve no idea where to turn next.  Diginet?  If only it was as good as they claim, and if only it was affordable.

3 Responses to “Broadband … really?”

  1. Malcolm says:

    Personally, I’ve had very few issues with my WiMAX connection from Telkom using an IS dsl fibre account.

    Here’s some ping stats from my side to and These were done without closing any of my existing connections and with my laptop also on the network.

    — PING Statistics—-
    200 packets transmitted, 200 packets received, 0.0% packet loss
    round-trip (ms) min/avg/max/med = 35/60/94/59

    — PING Statistics—-
    200 packets transmitted, 200 packets received, 0.0% packet loss
    round-trip (ms) min/avg/max/med = 304/329/375/329

    Granted, this doesn’t exactly reflect the high volumes of VoIP traffic and it was only 200 packets per ping (running concurrently through), but it does show the stability of this technology. A flood ping still doesn’t have any packet loss, but local traffic slowed down over time. This was either from strain on the link, or from something between me and that was slowing it down due to the high volumes. This flood took about 4 seconds for local and 5.5 seconds for international to complete.

    — PING Statistics—-
    1000 packets transmitted, 1000 packets received, 0.0% packet loss
    round-trip (ms) min/avg/max/med = 36/237/434/239

    — PING Statistics—-
    1000 packets transmitted, 1000 packets received, 0.0% packet loss
    round-trip (ms) min/avg/max/med = 309/418/595/394

    Good luck though applying for WiMAX in an area with DSL coverage, although, I think if a business pushes enough, they can get it.

    But it still doesn’t show any better latency than you are seeing.

  2. Jaco Kroon says:

    No, DSL is definitely the most stable thing out there when it works well. And relatively cheap too when looking at larger quantities of bandwidth. The only real issue I have with ADSL is the line-rental fee. The rest I’m fine with. I must admit that since I wrote this things have been much better in the stability department, except for one IS router which they claimed was a congested DSL exchange/line they couldn’t do anything about … changed the account to a SAIX based one and the issues I’ve had with that line just went away.

  3. Malcolm says:

    WiMAX = no line rental, R240 p/m excl access portion for 512k