Search results

:)

Straight down, straight up is always a polling problem. If there was an outage, logically, traffic would gradually increase after the dip, not ramp straight up to the same level. The graph you linked is SNMP based, we actually use sflow data internally, and our intention is to rid the entire...

It is a bit unfortunate that we are seeing these numbers for the worst reasons, but if anyone is interested, you can follow our stats here: http://ix.nap.africa/statistics/ixp We also have a live map of our JHB infrastructure here: https://www.napafrica.net/technical/napafrica-topology/

There isn't really a short answer, but you've mentioned a key point: Shorter path, but also control as a network over how your packets get to their destination. Since there is no charge to peer at NAPAfrica, and port sizes are virtually unlimited, you can also factor in cost reduction. No, you...

Correct, local servers require users. I think we're in a bit of a chicken-egg situation in that gamers want to compete against "the best", and the perception is that that means playing internationally. Unfortunately, you cannot change the physics of latency, so we should rather focus on making...

The "connected networks" are Autonomous Sytems Without making it sound too boring, NAPAfrica is infrastructure that allows networks to exchange traffic at scale at (typically) no cost.

Keep in mind that we operate at layer 2. Routing at layer 3 is controlled by connected networks, so we cannot influence where traffic is routed.

Indeed, this was an issue on our side in Johannesburg. We had an issue at 20:23 which has since been resolved. Apologies everyone!

This one is on us. We experienced an issue in Johannesburg at around 20:23, since resolved.

As previously mentioned, those particular stats are snmp based, which is simply not reliable at scale.

It is why many RRD based graphs end with "now-10m" so that you don't see the "in progress" bits. The problem with RRD aggregation is that it is space efficient, but you lose all granularity. We have moved most stats to a combination of sflow->influxdb and gNMI->Prometheus with 10s intervals.

Sure, so as posted, it is just timing. The jump would never be "straight up". The image also showed all regions dropping to zero. This would have affected around 430 networks. I'm sure you would have seen a few more complaints.

As mentioned, it is just a timing thing. With SNMP, straight down, straight up usually points to a collection problem. A full outage would be straight down, gradually tapering up. What you saw in the image was mid-collection. i.e. There was no data for that point in time. When refreshed, the...

Nothing of the sort. You will occasionally see that dip at the tail while data is being collected. It is unfortunately one of the downfalls of SNMP and the 5 minute aggregation, which we are gradually replacing. If you want to see live data on the same website, click "Technical" -> "NAPAfrica...