| *** diablo_rojo has joined #openstack-tc | 00:02 | |
| *** eandersson has quit IRC | 00:16 | |
| *** eandersson_ has joined #openstack-tc | 00:17 | |
| *** tosky has quit IRC | 00:23 | |
| *** cdent has quit IRC | 00:39 | |
| zaneb | fungi: a 'Procedural' label in Gerrit is not a horrible idea | 01:11 |
|---|---|---|
| *** jaosorior has quit IRC | 01:20 | |
| *** jamesmcarthur has joined #openstack-tc | 01:23 | |
| *** jaosorior has joined #openstack-tc | 01:32 | |
| fungi | we've discussed that possibility some in the past. i don't remember why we didn't do it | 01:34 |
| *** mriedem is now known as mriedem_afk | 01:39 | |
| *** jamesmcarthur has quit IRC | 01:49 | |
| *** whoami-rajat has joined #openstack-tc | 02:02 | |
| *** mriedem_afk has quit IRC | 02:24 | |
| *** jamesmcarthur has joined #openstack-tc | 02:25 | |
| *** mriedem has joined #openstack-tc | 02:29 | |
| openstackgerrit | Hongbin Lu proposed openstack/governance master: Change Zun PTL email address https://review.openstack.org/639242 | 02:38 |
| *** lbragstad has quit IRC | 02:44 | |
| *** mriedem has quit IRC | 03:12 | |
| *** jamesmcarthur has quit IRC | 03:25 | |
| *** jamesmcarthur has joined #openstack-tc | 03:28 | |
| *** jamesmcarthur has quit IRC | 03:36 | |
| *** jamesmcarthur has joined #openstack-tc | 03:40 | |
| *** jamesmcarthur has quit IRC | 03:45 | |
| *** jamesmcarthur has joined #openstack-tc | 04:07 | |
| *** jamesmcarthur has quit IRC | 04:23 | |
| *** lbragstad has joined #openstack-tc | 04:53 | |
| *** Luzi has joined #openstack-tc | 06:54 | |
| *** logan- has quit IRC | 07:35 | |
| *** logan- has joined #openstack-tc | 07:37 | |
| *** lbragstad has quit IRC | 07:44 | |
| *** Luzi has quit IRC | 08:18 | |
| *** bsilverman has quit IRC | 08:19 | |
| *** Luzi has joined #openstack-tc | 08:32 | |
| *** tosky has joined #openstack-tc | 08:39 | |
| *** dtantsur|afk is now known as dtantsur | 08:46 | |
| *** jpich has joined #openstack-tc | 08:53 | |
| *** dklyle has quit IRC | 09:25 | |
| *** e0ne has joined #openstack-tc | 09:25 | |
| *** dklyle has joined #openstack-tc | 09:27 | |
| *** tbarron has quit IRC | 09:30 | |
| *** cdent has joined #openstack-tc | 09:43 | |
| *** ricolin has joined #openstack-tc | 10:10 | |
| *** jamesmcarthur has joined #openstack-tc | 11:44 | |
| *** Luzi has quit IRC | 11:54 | |
| *** Luzi has joined #openstack-tc | 11:55 | |
| mugsie | fungi: zanebwe kind of did - http://git.openstack.org/cgit/openstack-infra/project-config/commit/gerrit/acls/openstack/designate.config?id=fdb962758c24698c89fe2ac7bc27182697c35d40 | 12:03 |
| mugsie | I think cinder did something similar | 12:03 |
| mugsie | or maybe it was glance | 12:03 |
| fungi | oh, good point! i guess designate-release-manager members can vote Review-Priority=-1 to effectively block a change from merging | 12:05 |
| mugsie | yeah - it was part of my motivation for the patch | 12:12 |
| *** logan_ has joined #openstack-tc | 12:27 | |
| *** logan_ is now known as Guest11647 | 12:28 | |
| *** diablo_rojo_ has joined #openstack-tc | 12:30 | |
| *** gouthamr has quit IRC | 12:30 | |
| *** cdent has quit IRC | 12:30 | |
| *** logan- has quit IRC | 12:30 | |
| *** diablo_rojo has quit IRC | 12:30 | |
| *** openstackgerrit has quit IRC | 12:30 | |
| *** Guest11647 is now known as logan- | 12:31 | |
| *** gouthamr has joined #openstack-tc | 12:33 | |
| smcginnis | mugsie: I think both projects have now. | 12:33 |
| *** jamesmcarthur has quit IRC | 12:34 | |
| mugsie | yeah, I know jokke pretty interested in it | 12:34 |
| *** jamesmcarthur has joined #openstack-tc | 12:34 | |
| *** mriedem has joined #openstack-tc | 12:50 | |
| *** jamesmcarthur has quit IRC | 12:52 | |
| *** jamesmcarthur has joined #openstack-tc | 12:57 | |
| *** cdent has joined #openstack-tc | 13:30 | |
| *** jamesmcarthur has quit IRC | 13:35 | |
| *** dtantsur is now known as dtantsur|afk | 13:38 | |
| *** jamesmcarthur has joined #openstack-tc | 13:39 | |
| *** jamesmcarthur has quit IRC | 13:40 | |
| *** jamesmcarthur has joined #openstack-tc | 13:40 | |
| *** zaneb has quit IRC | 14:04 | |
| *** zaneb has joined #openstack-tc | 14:07 | |
| mnaser | FYI. I think our community should be involved in cnf testbed. | 14:26 |
| mnaser | making sure that cloud is properly configured | 14:27 |
| mugsie | mnaser: yeah, that seems like a good colab opportunity | 14:29 |
| fungi | agreed, i'd hate to assume the folks running that are more interested in kubernetes vs openstack politics than getting accurate results both projects can use to identify opportunities for improvement | 14:29 |
| mugsie | I have found that the k8s vs openstack is usually a subset of people, who don't really do anything useful, and people that sign up for things like this work tend to be realists who just want things to be better | 14:30 |
| fungi | yep, also the people who designed and built the testbed are almost certainly not the ones responsible for the press release | 14:31 |
| fungi | so i wouldn't want to blame them for fanning the flames | 14:31 |
| *** lbragstad has joined #openstack-tc | 14:40 | |
| dims | fungi : mnaser : mugsie : ack on all counts. there's an opportunity if someone is interested. they explicitly call out for "improvements to either style deployments" in their presentation - https://docs.google.com/presentation/d/1nsPINvxQwZZR_7E4mAzr-50eFCBhbCHsmik6DI_yFA0/edit#slide=id.g5036f143e9_3_678 | 14:44 |
| mnaser | dims, fungi, mugsie: i've tried to help out to as much as i can in the openstack deployment, but, there's def not as much openstack chops on that team as k8s and i wish we can fix that | 14:45 |
| mnaser | but i've done my best to help out in terms of time | 14:45 |
| mugsie | dims: great | 14:45 |
| dims | thanks mnaser! | 14:46 |
| mnaser | what i suggested is they switch to openstack-ansible | 14:46 |
| mnaser | jamesdenton has been working on adding vpp support to OSA | 14:46 |
| mnaser | which means that the hacked up combination of chef code can be replaced by a properly tested project | 14:46 |
| dims | mnaser : slide 39 says terraform/ansible | 14:47 |
| mnaser | dims: yeah its terraform ansible to drive chef afaik | 14:47 |
| dims | aha | 14:48 |
| mnaser | terraform for packet, ansible for initial init and to run chef | 14:48 |
| mnaser | dims: | 14:48 |
| mnaser | https://github.com/cncf/cnf-testbed/blob/master/tools/deploy_openstack_cluster.sh#L52 | 14:48 |
| mnaser | ansible playbook inside docker to isntall openstack via chef | 14:49 |
| dims | gotcha | 14:49 |
| mnaser | :P | 14:49 |
| mnaser | that | 14:49 |
| mnaser | that's a lot of them | 14:49 |
| mnaser | tech, jesus, i am struggling to type | 14:49 |
| mnaser | not a good time for wrist problems in the current election eek | 14:49 |
| dims | LOL | 14:49 |
| fungi | i have to assume cnf performance with kubernetes on openstack bare metal instances would perform basically the same as the terraform-managed servers at packethost. it's too bad even the readme in that repo wants to conflate "openstack" with "virtual machines" | 14:52 |
| fungi | it doesn't seem like they're really measuring performance of "openstack" or "kubernetes" anyway, so it's odd that they frame it that way. the apis used to manage the containers or virtual machines don't really factor into it, do they? | 14:55 |
| fungi | i do find it reassuring that the cncf has the same git repository namespace confusion problems we do... the readme on https://github.com/cncf/cnf-testbed/ includes the sentence "The CNF Testbed is not a CNCF-hosted project." | 15:06 |
| *** jamesmcarthur has quit IRC | 15:11 | |
| *** jamesmcarthur has joined #openstack-tc | 15:11 | |
| fungi | ahh, i guess they're also testing/comparing time to deploy the infrastructure and time to deploy a single vnf/cnf | 15:14 |
| smcginnis | Kind of an apples to oranges comparison for the first part there. | 15:15 |
| *** jamesmcarthur has quit IRC | 15:15 | |
| fungi | sure. the more interesting benchmarks are still the network throughput when shuffling packets between userspace and kernelspace vs doing it all in userspace and/or directly passing the packet payloads between namespaces | 15:16 |
| *** e0ne has quit IRC | 15:16 | |
| *** Luzi has quit IRC | 15:17 | |
| smcginnis | Yeah, that is the more interesting part. | 15:17 |
| fungi | though what i think it's all missing is that the entire point of nfv in the first place was to be able to spread these functions across commodity hardware with conventional interconnectivity. as soon as your performance relies on all your functions being colocated on the same machine you're back to the expensive big-iron solutions telecommunication companies want to get away from in the first place | 15:18 |
| dhellmann | the speed of light is a pain | 15:19 |
| fungi | *but* for cases where you can couple those functions more closely on the same physical machine, it's certainly an efficiency gain | 15:19 |
| smcginnis | Still some benefit getting to one or a few big boxes. Back in my operator days, it's amazing to think about how much rack space was taken up by dedicated devices that really just moved data from one transport to another. | 15:21 |
| fungi | oh, absolutely. we usually ended up with upwards of a third of a rack devoted to the switches, firewalls, load balancers, et cetera for a one-rack deployment | 15:22 |
| jroll | right, telcos are moving NFV stuff further and further from the large datacenters. cell tower sites have very little space for computers. half height racks in many cases. | 15:22 |
| jroll | there, the benefit of NFV is that you can co-locate the specialized stuff (NFV) with other applications on the same commodity hardware | 15:23 |
| fungi | yep, and for the containerized network functions, there's at least some efficiency gained by dropping the additional context switching and hypervisor overhead, so can eke a bit more out of the same hardware or shrink your footprint a bit | 15:25 |
| jroll | indeed | 15:26 |
| *** jamesmcarthur has joined #openstack-tc | 15:27 | |
| fungi | their "pipeline" model with the direct data transfer between namespaces also drops a fair amount of protocol overhead. i bet for streams it's super efficient since you don't need to break at any particular frame size | 15:30 |
| *** jamesmcarthur has quit IRC | 15:30 | |
| fungi | oh, and also a biggie... the vnf images are coming from various vendors with specific kernel requirements. convincing them all to standardize on one kernel version and configuration is likely to be the biggest challenge for doing the same in containers | 15:35 |
| jroll | oh yes, great point | 15:37 |
| smcginnis | That's a really good point. | 15:37 |
| cdent | I've reports that many vnf vendors are miles away from being ready for containers | 15:39 |
| fungi | heck, you can have one environment interconnecting vnfs built on gnu/linux, netbsd and windows if you need | 15:39 |
| *** jamesmcarthur has joined #openstack-tc | 15:39 | |
| fungi | cdent: yeah, aiui, in a lot of cases the vendors have struggled just to get their operating systems to run on a virtual machine at all | 15:40 |
| cdent | yeah | 15:41 |
| fungi | rewriting them to rely on a mainline linux kernel abi? that's going to be a while | 15:41 |
| fungi | though i could see this research being useful for large telecommuncations providers who want to get into writing all their own network functions from the ground up | 15:42 |
| * jroll would be surprised if such a provider existed :) | 15:42 | |
| * cdent starts up a nascent huge telco | 15:44 | |
| * cdent cashes out | 15:44 | |
| fungi | i bet there's at least one provider with effectively unlimited labor and no need to respect intellectual property treaties who might see some benefit to giving it a go | 15:44 |
| fungi | given the main blockers are probably patent licensing and developers | 15:45 |
| clarkb | worth noting that their 65 minute openstack deployment time includes27.5 minutes of packethost provisioning | 15:48 |
| clarkb | not sure that that is actually what you want to be measuring since your baremetal provisioning should be a constant between the systems. | 15:49 |
| fungi | yeah, i think that's where mnaser's suggestion of switching them to osad comes in | 15:50 |
| fungi | where did you see the 27.5 minutes of packethost provisioning mentioned? | 15:52 |
| *** ricolin has quit IRC | 15:52 | |
| clarkb | https://github.com/cncf/cnf-testbed/issues/110#issuecomment-464230385 | 15:52 |
| *** e0ne has joined #openstack-tc | 15:53 | |
| fungi | cool, i hadn't thought to dig into the issues for that repo yet | 15:53 |
| clarkb | https://github.com/cncf/cnf-testbed/issues/111#issuecomment-464228115 does not include packethost deploy as far as I can tell | 15:53 |
| clarkb | so its an unfair comparison there as well | 15:54 |
| clarkb | (but also the data there is really scant and people aren't using a lot of words to go with the numbers so who knows) | 15:54 |
| fungi | https://github.com/cncf/cnf-testbed/issues/180#issuecomment-463604216 seems to imply it's steps 1 and 2 in that second bubble (so everything before chef kicks in) | 15:56 |
| fungi | or maybe it's just step 1 there | 15:57 |
| clarkb | step 4 is the packethost deployment step | 15:57 |
| clarkb | what I would actually find interesting is whether or not 1.1 million pps is expected to be a good number for the virtual machine case | 15:58 |
| clarkb | probably need to talk to the fd.io folks for that | 15:58 |
| fungi | zaneb: your "convenient fiction" of representative government is right on the money. nice answer! | 17:06 |
| cdent | The whole world is a convenient fiction, which I find incredibly inconvenient. | 17:09 |
| cdent | Except for Belgium, of course. | 17:09 |
| fungi | i hear belgium is extremely convenient, but i haven't been (yet!) | 17:14 |
| *** jpich has quit IRC | 17:17 | |
| *** e0ne has quit IRC | 17:24 | |
| mugsie | clarkb: the number thrown around for ovs in Verizon was 200k pps, but I have no idea where that came from, and neither did anyone else I asked there | 17:31 |
| *** bsilverman has joined #openstack-tc | 17:55 | |
| zaneb | fungi: Belgium is very cool actually. try to swing a speaking slot or something at FOSDEM some time | 18:37 |
| fungi | i wanted to go this year but it was the same week as a vacation i'd planned nearly a year in advance. hoping 2020 will be the year i finally see fosdem (and belgium for that matter) | 18:38 |
| mugsie | FOSDEM is ++ and evrardjp is a handy local guide | 18:51 |
| fungi | so i gather | 19:06 |
| zaneb | dress warm though | 19:11 |
| mugsie | It was ok this year, just a little bit of snow. It was no Dublin | 19:13 |
| zaneb | I went in 2012 and it was -15C | 19:17 |
| smcginnis | Warmer than here. | 19:18 |
| *** gouthamr has quit IRC | 19:20 | |
| *** gouthamr_ has joined #openstack-tc | 19:20 | |
| *** mriedem has quit IRC | 19:21 | |
| zaneb | oh yeah. I was in NZ this Feb and it was quite pleasant :D | 19:21 |
| *** mriedem has joined #openstack-tc | 19:22 | |
| *** e0ne has joined #openstack-tc | 19:26 | |
| mtreinish | zaneb: yeah, NZ at the end of Jan. was really nice too | 19:31 |
| mtreinish | going from -20C to 25C was quite nice :) | 19:32 |
| *** e0ne has quit IRC | 19:46 | |
| *** e0ne has joined #openstack-tc | 19:46 | |
| fungi | i would crack from the thermal stress | 20:06 |
| *** jamesmcarthur has quit IRC | 20:15 | |
| *** jamesmcarthur has joined #openstack-tc | 20:16 | |
| *** jamesmcarthur has quit IRC | 20:20 | |
| *** e0ne has quit IRC | 20:31 | |
| *** diablo_rojo_ has quit IRC | 20:35 | |
| zaneb | bauzas: I'm stealing "the reality wall" :) | 20:36 |
| *** diablo_rojo has joined #openstack-tc | 20:36 | |
| *** gouthamr_ is now known as gouthamr | 20:42 | |
| *** spsurya has quit IRC | 21:03 | |
| *** cdent has quit IRC | 21:05 | |
| *** cdent has joined #openstack-tc | 21:05 | |
| *** jamesmcarthur has joined #openstack-tc | 21:11 | |
| *** zbr|out has quit IRC | 21:27 | |
| *** e0ne has joined #openstack-tc | 21:28 | |
| *** eandersson_ is now known as eandersson | 21:40 | |
| *** whoami-rajat has quit IRC | 21:41 | |
| *** cdent has quit IRC | 21:58 | |
| *** e0ne has quit IRC | 22:03 | |
| *** jamesmcarthur has quit IRC | 22:49 | |
| *** jamesmcarthur has joined #openstack-tc | 23:02 | |
| *** jamesmcarthur has quit IRC | 23:09 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!