operations
operations copied to clipboard
openstreetmap
Switch Amsterdam to redundant 10Gbps connection
We currently have a 1Gbps connection at Equinix. We're nearing it's capacity
The biggest single user seems to be Apache on ironbelly
Volume is up significantly from 9 months ago, are we sending more traffic to ironbelly?
If we want to upgrade our network connection, do we want to go for 2x GigE, 1x 10 GigE, or 2x 10 GigE?
Saying "apache" isn't very helpful - it's more useful to know what machine you're talking about than what service.
Are you saying ironbelly is the source of the extra traffic?
Edited to add missing info.
Are you saying ironbelly is the source of the extra traffic?
Yes - it's got an order of magnitude more traffic than pre-move.
Indeed but I don't understand why...
It appears that grisu was carrying similar load when it was standing in during the move so the jump appears to have been in mid July at the point when load was transferred from ironbelly to grisu for the move.
I thought for a minute that it was when heanet stopped mirroring, and we stopped redirecting to them (see #210) but that was in May.
That said we're only talking about 150-200Mbit/s average, so about 20% of the capacity.
- I've asked HEnet (2x 10G) + Equinix Cross Connect (AM6/AM7) for quotes
- I've emailed AMS-IX asking if they can help or point us at anyone.
- I've asked Vorboss if they can quote. They are based in AM6. (via 2x 10G cross connects)
Can we also ask Equinix about the prices for a connection inside the same DC? The difference might not be prohibitive.
Let's also ask other ISPs in AM6 if we can't get a satisfying response.
https://www.peeringdb.com/fac/2906
When we do this, we should also get two physical uplinks (one for each switch) for redundancy - see recent outage where one switch was dead and the other was fine but had no uplink.
Cross-connections between AM6 and AM7 HE are about the same cost as within AM6. I propose moving ahead with HEnet. The other option would be starting to do BGP stuff.
Let's reassess our capacity planning after #678 - ironbelly will use a lot less bandwidth with planet files on S3.
HE Diverse Cross Connects are now connected. Patched across to the 2x switches. Network config to follow.
We have higher than expected signal loss on a new cross connect to HE. It is being investigated by the team at Equinix.
The issue with higher than expected signal loss has now been resolved. The issue was a faulty patch cable on remote end.
Blocked by decision on Private ASN BGP ($$$?) or static IP subnet routing.
"So, we hereby confirm that service will be ceased by 1st August 2023." - Cogent