Pim van Pelt
d17e1c62e5
4.4 KiB
title, date
| title | date |
|---|---|
| Week 14, Tuesday: Patches Galore | 2024-10-29T21:55:00+02:00 |
{{< image frame="true" width="17em" float="right" src="/img/headline/dreamscape-03.png" alt="Credit: Dreamscape, Kush Sessions, YouTube" >}}
This morning my buddy Arend sends me a message on Telegram - and asks me if I can check the port in Qupra. Oh my deity, it's finally happening! After the general assembly of Coloclue approved a member's petition to allow members to install cross connects at Qupra, a few months of "kastje, muur" happened, and the networking committee and association board reached an agreement on how this would happen.
A few months after that, Arend tried to install a crossconnect but we weren't ready paper-wise. I brought it back to the attention of the networking committee and we identified the missing pieces: a patch panel with keystones needed to be installed in each rack (and it was only available in two of the racks at the time), and a change to the administrative database needed to be added to document which members had which crossconnects.
Tim took care of the first thing - he ordered the panels and keystones and went to the datacenter to install them. I offered to take careo f the second thing - but since the administrative database has need-to-know information, our treasurer Arjan preferred to add the records himself. Once these two things were taken care of, all I had to do is wait for a practical moment :) I had planned to deploy the fiber myself last week, but I had to cancel my trip to Amsterdam due to the COVID situation.
So I was surprised and delighted that Arend pinged me. The Qupra FrysIX switch was pre-configured, and all that was left was to plug things in. Arend made quick work of it, and as well put in the cross connect for a few other members at Coloclue, he's such a sweetheart! For me, this link will be used to alleviate the hypervisor at Equinix AM3, as it is running low on disk throughput due to me using Samsung consumer SSDs. I shipped Arend a few enterprise SAS SSDs before, but he hasn't gotten around to deploying them yet. More importantly, the AM3 hypervisor runs FrysIX routeserver, LibreNMS and IXPManager.
After the Qupra gig, Arend made his way to NIKHEF where he installed the FrysIX patch for FreeIX
Remote, directly into the VPP router nlams0.net.free-ix.net. That router now has LSIX, SpeedIX,
and FrysIX connected. I spend some time bringing FreeIX Remote AS50869 into quarantine and then into
the production VLAN. That's a benefit of running the IXP: I get to expedite my own connections :)
Now that the FreeIX Remote router is connected to FrysIX, I allocate a private VLAN between it and
IPng's infrastructure. This allows me to create a VPWS (L2VPN, Ethernet over MPLS) on IPng's MPLS
switches msw0.nlams0 and msw1.chrma0 from this router in Amsterdam to the one I already installed
in Zurich. iBGP comes up, and there are now three routers in play (nlams0, chrma0, and
grskg0), and amongst them, they know about 207K IPv4 prefixes and 64.7K IPv6 prefixes, and all of
them can be reached via direct or routeserver peering. How cool is that?
pim@nlams0:~$ birdc show route count
BIRD v2.15.1-4-g280daed5-x ready.
800934 of 800934 routes for 207438 networks in table master4
449754 of 449754 routes for 64696 networks in table master6
1501107 of 1501107 routes for 500369 networks in table t_roa4
364077 of 364077 routes for 121359 networks in table t_roa6
Total: 3115872 of 3115872 routes for 893862 networks in 4 tables
In the evening I send a maintenance announcement out to FrysIX members: in the night of Wednesday to Thursday of this week, I will move the routeserver RS2 and the IXPManager over to the hypervisor at Qupra, which now sports a 10G connection to the FrysIX peering switch there. I have plumbed the management VLAN 264 and the Quarantine 2605 and the Peering LAN 2604 through to the hypervisor.
I practice by moving nms.frys-ix.net over - this is a non-intrusive change. Using ZFS block device
replication, I can pump over the boot disk with about 110MB/s, because the hypervisor itself has
"only" a one gigabit connection. I boot the VM, and it comes up cleanly. Nice. I spend a few hours
preparing the move of the other two machines (RS2 and IXPManager), which are service impacting. But
I can start by making a snapshot of the block devices, copy their data over ahead of time, and then
copy a final snapshot incrementally.
Today was a good day for FrysIX :)