I would request for IPv6 support as well. Just sent email to that ID to open a discussion about it. It’s critical to support it now as IPv4 performance at times is degraded on eyeball networks due to massive CGNATs and other issues.

The grafana graph is interesting. Assuming same ratios of inbound/outbound is there overall and not just on the peering link gives an impression that backblaze is inbound heavy. I always felt it would be outbound heavy (as most of hosting, object storage players are). This seems like the case that backup offering is still larger than B2 traffic volume wise and hence more inbound traffic. Is that true?

Equinix IX in San Jose is the latest addition — we’ve been expanding our peering over the years and we’ve had connectivity to AMS-IX in Amsterdam, Any2East in Reston, and Ninja-IX in Phoenix for a long time. Finding a IX that has local DC or campus connectivity with mutual traffic benefits is the challenge. Sometimes all the puzzle pieces don’t align. We would love to peer with everyone, but we have to be strategic about it.

We do have sflow monitoring at our edge and use that with enriched BGP data (AS, AS total path, flow size) to help make decisions about peering and traffic engineering.

We peer with the Equinix Route Servers and a few peers that we’ve identified as high priority or high impact to our traffic flow. So we are learning any route that a member is sending to the Route Servers and routes from our one-to-one peers. Yep, we’re moving more and more into automation for peer deployment.

Thanks for the comments! I think some of the topics here could become full blog posts that we could write about. I’ll keep these in mind for future content!

Thanks for the feedback, we’re aware of the demand for IPv6 and we’re actively working on it. We’ll have more to report on this soon!

I would guess you are using server logs vs something similar to netflow. Generating and analyzing netflow from the routers at your traffic levels would be an interesting challenge.

I see you are an open peer on peeringdb, nice. Are you peered with everyone at SV1 or just reaching out to the high traffic ones and any peering requests? That’s a decent amount of BGP peers if you aren’t template-ing them and well the administrative cost of maintaining all the peers / contacting networks when something goes wrong.