| *** rvasilets___ has quit IRC | 00:29 | |
| *** arnoldje has joined #openstack-performance | 00:38 | |
| *** harshs has joined #openstack-performance | 00:59 | |
| *** harshs_ has joined #openstack-performance | 01:05 | |
| *** harshs has quit IRC | 01:07 | |
| *** harshs_ is now known as harshs | 01:07 | |
| *** harlowja has quit IRC | 01:38 | |
| *** harlowja_ has joined #openstack-performance | 01:38 | |
| *** badari has quit IRC | 02:12 | |
| *** harshs has quit IRC | 03:32 | |
| *** harshs has joined #openstack-performance | 03:35 | |
| *** harshs has quit IRC | 03:53 | |
| *** badari has joined #openstack-performance | 05:34 | |
| *** harshs has joined #openstack-performance | 05:41 | |
| *** arnoldje has quit IRC | 06:03 | |
| *** badari has quit IRC | 06:18 | |
| *** harshs has quit IRC | 06:32 | |
| *** harshs has joined #openstack-performance | 06:41 | |
| *** harshs has quit IRC | 06:52 | |
| *** beisner has quit IRC | 07:44 | |
| *** beisner has joined #openstack-performance | 07:45 | |
| DinaBelova | regXboi - either me or Boris will do that | 08:38 |
|---|---|---|
| DinaBelova | notmorgan - not sure what's time in your TZ :) it's 3PM UTC :) | 08:38 |
| notmorgan | same as spamaps | 08:40 |
| * notmorgan lives on the west coast of the US | 08:41 | |
| notmorgan | LA for now, Portland soon ;) | 08:41 |
| DinaBelova | notmorgan :) | 08:45 |
| *** amaretskiy has joined #openstack-performance | 09:26 | |
| DinaBelova | harlowja_, SpamapS, Kristian_, regXboi, patrykw_ - please take some time on reviewing https://review.openstack.org/#/q/status:open+project:openstack/rally+branch:master+topic:bp/vm-workloads-framework,n,z workloads changes for rally, that will allow to introduce data plane testing | 09:29 |
| boris-42 | OO already 20 people in chat nicceeeee | 09:32 |
| kun_huang | boris-42: 31 people tomorrow ;) | 09:32 |
| boris-42 | Finally chat where I can say that OpenStack doesn't scale well and perform bad and it will be ok | 09:32 |
| boris-42 | =) | 09:33 |
| *** wootehfoot has joined #openstack-performance | 10:47 | |
| *** dburnazyan has quit IRC | 10:54 | |
| DinaBelova | boris-42 - you found the right place! :D | 11:29 |
| *** wootehfoot has quit IRC | 11:59 | |
| *** mwagner has quit IRC | 12:08 | |
| *** nihilifer has joined #openstack-performance | 13:28 | |
| *** arnoldje has joined #openstack-performance | 13:29 | |
| *** mwagner has joined #openstack-performance | 13:42 | |
| *** arnoldje has quit IRC | 14:45 | |
| *** arnoldje has joined #openstack-performance | 15:20 | |
| *** redixin has joined #openstack-performance | 15:30 | |
| *** badari has joined #openstack-performance | 15:34 | |
| *** wootehfoot has joined #openstack-performance | 15:50 | |
| *** harshs has joined #openstack-performance | 16:13 | |
| *** kun_huang_ has joined #openstack-performance | 16:13 | |
| *** boris-42_ has joined #openstack-performance | 16:13 | |
| *** harshs has quit IRC | 16:14 | |
| *** notmorga1 has joined #openstack-performance | 16:16 | |
| *** kun_huang has quit IRC | 16:17 | |
| *** notmorgan has quit IRC | 16:17 | |
| *** boris-42 has quit IRC | 16:17 | |
| *** kun_huang_ is now known as kun_huang | 16:17 | |
| *** notmorga1 is now known as notmorgan | 16:18 | |
| *** boris-42_ is now known as boris-42 | 16:19 | |
| *** harshs has joined #openstack-performance | 17:24 | |
| *** harshs has quit IRC | 17:37 | |
| *** wootehfoot has quit IRC | 17:38 | |
| *** amaretskiy has quit IRC | 17:43 | |
| *** harlowja_ has quit IRC | 18:08 | |
| *** boris-42 has quit IRC | 18:28 | |
| *** harlowja has joined #openstack-performance | 18:42 | |
| *** harlowja_ has joined #openstack-performance | 18:44 | |
| *** harlowja has quit IRC | 18:44 | |
| harlowja_ | DinaBelova hey, ya i saw the workflow stuffs | 19:01 |
| harlowja_ | haven't investigated to much, looks interesting | 19:01 |
| harlowja_ | it will i think be useful to determine the impact of overcommits on workloads | 19:01 |
| harlowja_ | and/or to convince people that vms are ok (vs containers, baremetal...) | 19:01 |
| harlowja_ | we've been trying to figure out a good way to determine what our CPU, ram... overcommit will be, and having such kind of workload tests will help determine the impact | 19:02 |
| harlowja_ | if say we move CPU overcommit to 2x vs 3x and ... what will the result on 'real' workflows be... | 19:02 |
| harlowja_ | *because once u turn a CPU overcommit to say 3x, its somewhat hard to go back (aka what do u do about hypervisors that now have to many VMs on them...) | 19:03 |
| harlowja_ | so tweaking overcommit has always been a process we (at yahoo) has been careful/cautious about... | 19:03 |
| harlowja_ | because it has real side-effects | 19:03 |
| harlowja_ | and then as a final goal of this, u can feed 'profiles' that users have of what there workload is back into the scheduling process and u can better place them | 19:06 |
| harlowja_ | ya da ya da | 19:06 |
| kun_huang | harlowja_: tuning "overcommit" as a parameter? | 19:41 |
| harlowja_ | kun_huang overcommit ratios | 19:41 |
| harlowja_ | 'cpu_allocation_ratio' | 19:42 |
| harlowja_ | 'disk_allocation_ratio' | 19:42 |
| harlowja_ | and other allocation ratios that nova uses | 19:42 |
| harlowja_ | how do people know what good values to change those to without knowing the effect on the workloads in there environment... | 19:42 |
| harlowja_ | 'ram_allocation_ratio' | 19:43 |
| harlowja_ | ^ has a bigger impact (because of the OOM killer) | 19:43 |
| kun_huang | got it, rally could deploy openstack now, by devstack | 19:43 |
| harlowja_ | right | 19:43 |
| kun_huang | and we could adjust those config there | 19:43 |
| harlowja_ | but that doesn't solve the 'what effect does this value have on workloads in my environment' problem | 19:43 |
| harlowja_ | *where workloads in each environment will be different | 19:44 |
| harlowja_ | ie, for yahoo, it could be a large amount of http server workloads... | 19:44 |
| harlowja_ | for other companies it might not be that... | 19:44 |
| harlowja_ | the scheduler should really have knowledge about what your 'expected workload' is and place it somewhere that best fits | 19:44 |
| kun_huang | I know this, it depends on real workloads | 19:44 |
| harlowja_ | so https://review.openstack.org/#/q/status:open+project:openstack/rally+branch:master+topic:bp/vm-workloads-framework,n,z is a good start to making something possible in a nice way | 19:45 |
| kun_huang | a lot of work need be fininshed in vm workload benchemark | 19:45 |
| harlowja_ | ya, its a good start i think | 19:45 |
| kun_huang | harlowja_: google has one | 19:45 |
| harlowja_ | i bet | 19:45 |
| kun_huang | wait a minute, I'm looking for my notes | 19:45 |
| harlowja_ | k | 19:45 |
| harlowja_ | SpamapS will likely want one also :-P | 19:45 |
| harlowja_ | smarter scheduling based on real profiles of real workloads would be super | 19:46 |
| kun_huang | we all need this | 19:46 |
| kun_huang | https://github.com/GoogleCloudPlatform/PerfKitBenchmarker | 19:46 |
| kun_huang | it looks like they use this framework to do SINGLE vm workload | 19:46 |
| harlowja_ | now make that dynamic smarter scheduling that adapts to workloads (aka a user doesn't even have to tell u there workload profile) and u win the million dollar prize | 19:46 |
| harlowja_ | ya, to me the artifical benchmarking stuff is only so useful | 19:46 |
| harlowja_ | dynamic workload balancing would be better, based on real workloads from real people/companies | 19:47 |
| harlowja_ | and thats where i think openstack has the oppurtunity to do that, beause we have real people/companies with diverse sets of workloads/profiles | 19:47 |
| kun_huang | get your ideas ;) | 19:47 |
| kun_huang | harlowja_: btw do you want to review those codes? | 19:47 |
| harlowja_ | sooo we (as a group/community) can do something that doesn't use artifical benchamrks/profiles... | 19:47 |
| harlowja_ | sure | 19:47 |
| harlowja_ | i'll try :-P | 19:47 |
| kun_huang | harlowja_: it's welcome | 19:48 |
| harlowja_ | yup | 19:48 |
| kun_huang | I have saw some design points in those patches | 19:48 |
| kun_huang | and part of them is hard to understand | 19:48 |
| kun_huang | there is a document, but no details | 19:49 |
| harlowja_ | ? | 19:51 |
| harlowja_ | document wheree? | 19:51 |
| kun_huang | harlowja_: https://docs.google.com/document/d/1FQicnj6aIdbfYts9ofYdz0dUEf9zyssWqDjQR40h268/edit#heading=h.2vn8vxb5ltcu | 19:53 |
| kun_huang | here is one | 19:53 |
| harlowja_ | cool | 19:53 |
| harlowja_ | thx | 19:53 |
| kun_huang | anyway, your ideas are welcome | 19:54 |
| kun_huang | I need go to bed now | 19:54 |
| *** mwagner has quit IRC | 20:11 | |
| SpamapS | harlowja_: I will likely want one what? | 20:22 |
| harlowja_ | the ability to determine based on workloads how to change overcommit ratios | 20:23 |
| harlowja_ | and eventually having the ability to automatically schedule/reschedule workloads based on dynamic workload info | 20:23 |
| SpamapS | harlowja_: FYI, IMO, smart scheduling is a _massive_ complexity problem that I don't actually think should be handled inside OpenStack. Flavors oriented toward workloads, and let users pick flavors that work for their workload. | 20:23 |
| harlowja_ | nothing worth doing is going to be easy ;) | 20:23 |
| harlowja_ | never is, ha | 20:23 |
| harlowja_ | but i get what u are saying | 20:24 |
| SpamapS | If we try too hard to make it magical, it will work like magic: only from one angle. | 20:24 |
| * SpamapS tweets that | 20:26 | |
| harlowja_ | lol | 20:27 |
| harlowja_ | SpamapS fair enough, but if u don't plan for magic, u will never learn how to do the magic in the first place | 20:28 |
| * harlowja_ doesn't have/use tweeter so can't tweet that | 20:28 | |
| harlowja_ | lol | 20:28 |
| harlowja_ | and then u will never be a magician | 20:31 |
| harlowja_ | obviously | 20:31 |
| SpamapS | damn | 20:38 |
| SpamapS | crushing my dreams | 20:38 |
| *** mwagner has joined #openstack-performance | 20:54 | |
| *** david-bravo23 has joined #openstack-performance | 23:27 | |
| david-bravo23 | (CEPH RBD Openstack) RBD_STORE_CHUNK_SIZE? Does anyone have any insight on what is the best for SQL (Many Small frequent read/writes)? Default 4MB, or a smaller size like 1MB? | 23:27 |
| *** boris-42 has joined #openstack-performance | 23:32 | |
| *** arnoldje has quit IRC | 23:37 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!