i have been running irrd for some years. am about to dump that (virtual) server and move from freebsd to bullseye. is there anything more modern, and _simpler_, than irrd at which i should take a look? randy
On Sat, Jun 18, 2022 at 2:37 PM Randy Bush <randy@psg.com> wrote:
i have been running irrd for some years. am about to dump that (virtual) server and move from freebsd to bullseye. is there anything more modern, and _simpler_, than irrd at which i should take a look?
While IRRd4 fails the simpler criteria, it passes the more modern criteria of your question. And since both irrd-legacy and RWS (RIPE) seem to not be maintained, I believe it to be the only code that can be labeled as modern. What I can say from the experience of running TC IRR (32k objects) is that you need lots of RAM and vCPUs, much more than IRRd 2.9/3.x. https://github.com/irrdnet/irrd Rubens
On Sat, Jun 18, 2022 at 3:22 PM Randy Bush <randy@psg.com> wrote:
At least 32GB RAM
It runs fine now with 16GB RAM, after a code change. It will also take much less RAM if you turn RPKI validation off. This is from TC at this moment, which mirrors 5.5 million records from other IRRs and do RPKI validation of its own objects: free -g total used free shared buff/cache available Mem: 31 12 5 6 13 11 Swap: 3 0 3
At least 4 CPU cores
We run with 2 CPU cores, and indeed you might want to use 3 or 4: top - 18:27:59 up 4:24, 1 user, load average: 2.24, 2.19, 3.04
At least 150GB of disk space (SSD recommended)
Possibly less than that, we provisioned 100GB and it uses half of it. The host system indeed uses SSD, but it hosts many more VMs. An iostat of the VM as of now: avg-cpu: %user %nice %system %iowait %steal %idle 21.59 0.00 2.97 0.43 0.00 75.01 Device tps kB_read/s kB_wrtn/s kB_dscd/s kB_read kB_wrtn kB_dscd (...) sda 135.40 1356.78 3650.10 0.00 21898122 58912032 0 Rubens
It will also take much less RAM if you turn RPKI validation off.
oh dear ghod. do i need to turn the dancing donkeys off too? "Make each program do one thing well. To do a new job, build afresh rather than complicate old programs by adding new "features"." -- ken thompson - unix philosophy a good side to a bit of economic contraction might be a side effect of code bloat and featuritis contraction. randy, who has a 32G laptop and runs an editor with an rss of ~100MB
It will also take much less RAM if you turn RPKI validation off.
oh dear ghod. do i need to turn the dancing donkeys off too?
"Make each program do one thing well. To do a new job, build afresh rather than complicate old programs by adding new "features"." -- ken thompson - unix philosophy
a good side to a bit of economic contraction might be a side effect of code bloat and featuritis contraction.
to be clearer, i now run a 4GB VM with irrd2, rancid, nfsen, and a wiki. so i will stick with irrd2. randy
On Sun, Jun 19, 2022 at 6:07 PM Randy Bush <randy@psg.com> wrote:
It will also take much less RAM if you turn RPKI validation off.
oh dear ghod. do i need to turn the dancing donkeys off too?
"Make each program do one thing well. To do a new job, build afresh rather than complicate old programs by adding new "features"." -- ken thompson - unix philosophy
a good side to a bit of economic contraction might be a side effect of code bloat and featuritis contraction.
to be clearer, i now run a 4GB VM with irrd2, rancid, nfsen, and a wiki. so i will stick with irrd2.
Yeap, we tried installing IRRd 4 in a similar VM that used to run TC and it was a catastrophic failure. Although it was 4.0beta2 and some changes we are made since (IRRd4 is now at 4.2.4), I don't think they were enough to change the outcome. Rubens
Hi Randy, On Sun, 19 Jun 2022 at 23:07, Randy Bush <randy@psg.com> wrote:
It will also take much less RAM if you turn RPKI validation off.
oh dear ghod. do i need to turn the dancing donkeys off too?
"Make each program do one thing well. To do a new job, build afresh rather than complicate old programs by adding new "features"." -- ken thompson - unix philosophy
a good side to a bit of economic contraction might be a side effect of code bloat and featuritis contraction.
to be clearer, i now run a 4GB VM with irrd2, rancid, nfsen, and a wiki. so i will stick with irrd2.
Are you looking for to set up just an “authoritative IRR source” (RGNET?), or to set up an instance which mirrors all the world’s IRRs? The latter option is quite memory heavy. If mirroring other databases is not the goal; an “auth only” IRRd v4 deployment will easily fit your VM alongside those other apps. Or perhaps you are interested to fund development of a modern lightweight version of the IRRd software? :-) Kind regards, Job
I've seen recently a trend where code is optimized for run time and memory consumption is a distant second consideration. I think this is a side-effect of the growth of big data, where you really do have to worry about your run time. Unfortunately this seems to have creeped into a lot of other types of coding where it doesn't make a much sense. For example, it's not uncommon to see data stored in hash tables which admittedly are fast to retrieve if you use the right hash table key. But since most programmers don't think through the overhead of the key portion of the table, you'll often see many hash tables each containing the exact same set of keys, each holding one data item (say a distance in a routing table). A more memory efficient method is to store a structure in the hash table containing all the values. That way they only pay the key "tax" once. I could vent about all sort of other sins like this. On Sat, Jun 18, 2022, 1:28 PM Randy Bush <randy@psg.com> wrote:
It will also take much less RAM if you turn RPKI validation off.
oh dear ghod. do i need to turn the dancing donkeys off too?
"Make each program do one thing well. To do a new job, build afresh rather than complicate old programs by adding new "features"." -- ken thompson - unix philosophy
a good side to a bit of economic contraction might be a side effect of code bloat and featuritis contraction.
randy, who has a 32G laptop and runs an editor with an rss of ~100MB
* randy@psg.com (Randy Bush) [Sat 18 Jun 2022, 19:39 CEST]:
i have been running irrd for some years. am about to dump that (virtual) server and move from freebsd to bullseye. is there anything more modern, and _simpler_, than irrd at which i should take a look?
What are you running irrd for? Have you considered irrd4? Simpler than running your own would be querying e.g. rr.ntt.net. -- Niels.
participants (5)
-
Forrest Christian (List Account) -
Job Snijders -
Niels Bakker -
Randy Bush -
Rubens Kuhl