2019-07-22 05:51:15 UTC - chris: HI Guys, I created the files ( __main__.py and xxx.py) into a zip file. `action` and `invoke` print `ok`. However, when i run `activation get`, why log shows error `ModuleNotFoundError: No module named 'keras'` Does anybody know ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563774675008600?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 05:53:40 UTC - Satwik Kolhe: See if this blog post by @James Thomas - <http://jamesthom.as/blog/2017/04/27/python-packages-in-openwhisk/> - helps you package your python function https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563774820008900?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 05:56:16 UTC - chris: thanks for your replying. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563774976009200?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 06:08:05 UTC - Roberto Santiago: @chris, took me a bit to figure out how to package up python actions with `virtualenv` and `zip`. James' tutorial is great. Let me know if you hit any roadblocks. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563775685009600?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 06:11:44 UTC - chris: @Roberto Santiago OK!!thank you...you guys are really awesome :heart_eyes: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563775904009800?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 08:12:08 UTC - chris: Finally......Error again :fearful: the log shows that `ModuleNotFoundError: No module named 'pyjokes'` ....... https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563783128010000?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 08:13:03 UTC - chris: But i did installed it in `virtualenv` https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563783183010200?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 08:13:35 UTC - chris: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563783215010400?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 08:14:04 UTC - chris: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563783244010800?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 08:15:48 UTC - chris: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563783348011600?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 08:17:04 UTC - chris: Does anyone know why it not work for me !!!!!!!!!:dizzy_face: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563783424012000?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 08:51:27 UTC - Satwik Kolhe: Try, change values according to your environment
zip -r jokes.zip venv/bin/activate_this main.py venv/<path to site packages>/jokes 'activate_this' file is important https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563785487012200?thread_ts=1563774675.008600&cid=C3TPCAQG1 ---- 2019-07-22 09:25:14 UTC - Pepi Paraskevoulakou: hello anyone experienced with test: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563787514012700 ---- 2019-07-22 09:25:17 UTC - Pepi Paraskevoulakou: WARNING: The DOCKER_COMPOSE_HOST variable is not set. Defaulting to a blank string. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563787517012900 ---- 2019-07-22 09:25:33 UTC - Pepi Paraskevoulakou: need to do something specific? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563787533013300 ---- 2019-07-22 10:26:52 UTC - Michael Schmidt: ^This is a lack of experience with Docker Compose, I'd read up on docker compose itself. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563791212013900 ---- 2019-07-22 11:33:56 UTC - Roberto Santiago: Looking for some testing advice. I am having success with using triggers to orchestrate actions to fulfill functional requirements. While I am testing the actions individually (i.e. unit testing), I am wondering how to go about testing the orchestration of those actions, sort of like an integration test. Any thoughts? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563795236016300?thread_ts=1563795236.016300&cid=C3TPCAQG1 ---- 2019-07-22 11:49:54 UTC - Satwik Kolhe: How are triggers different from APIs exposed using openwhisk/apigateway ? Except that triggers just invoke a function without returning the actual result! https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563796194016400?thread_ts=1563795236.016300&cid=C3TPCAQG1 ---- 2019-07-22 12:08:42 UTC - Roberto Santiago: Triggers are like events. Many rules can be attached to a single trigger. Each of these rules invoke an action. But all of those invocations are independent of one another. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563797322016600?thread_ts=1563795236.016300&cid=C3TPCAQG1 ---- 2019-07-22 12:29:04 UTC - James Thomas: If you expose those actions as an API - I’d write e2e tests with those API endpoints. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563798544016800?thread_ts=1563795236.016300&cid=C3TPCAQG1 ---- 2019-07-22 12:29:22 UTC - James Thomas: otherwise, write e2e tests and set up and fix up the test environment and wait for the results https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563798562017000?thread_ts=1563795236.016300&cid=C3TPCAQG1 ---- 2019-07-22 16:31:12 UTC - Jona: Hey there ! I just discovered Apache OpenWhisk, and stumbled upon this on the homepage: ``` Scaling Per-Request & Optimal Utilization Run your action ten thousand times in a fraction of a second, or once a week. Action instances scale to meet demand as needed, then disappear. Enjoy optimal utilization where you don't pay for idle resources. ``` I wonder if someone could give more details on the kind of (hardware) setup needed to achieve *per-request scaling* while *don't pay for idle resources* ? I can in fact imagine installing OpenWhisk on an Amazon EC2 instance but I would still need to pay for idle resources... am I missing something ? :slightly_smiling_face: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563813072022500 ---- 2019-07-22 16:58:16 UTC - Michael Schmidt: You would use openwhisk as an on prem solution probably, or openwhisk-as-a-service https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563814696022800 ---- 2019-07-22 16:58:37 UTC - Michael Schmidt: AWS lamba exists, but I haven't heard tons of people crazy about it... https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563814717023300 ---- 2019-07-22 16:58:51 UTC - Michael Schmidt: that would be your aws get charged by the function solution though... https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563814731023700 ---- 2019-07-22 16:59:07 UTC - Michael Schmidt: you probably woun't put OpenWhisk on aws instances imo https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563814747024300 ---- 2019-07-22 17:02:02 UTC - James Thomas: <vendor-plug>If you want to just use “openwhisk-as-a-service” with those features check out IBM Cloud Functions</vendor-plug> whisking : Upkar Lidder, Michael Schmidt https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563814922025000 ---- 2019-07-22 17:02:38 UTC - Michael Schmidt: lol https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563814958025400 ---- 2019-07-22 17:02:43 UTC - Michael Schmidt: this too https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563814963025600 ---- 2019-07-22 17:39:02 UTC - Jona: AHahahah https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817142026100 ---- 2019-07-22 17:40:06 UTC - Jona: Was the homepage claims actually a <vendor-plug></vendor-plug> thing ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817206027000 ---- 2019-07-22 17:40:37 UTC - Jona: AWS EC2 was just an illustration https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817237027600 ---- 2019-07-22 17:40:39 UTC - Rodric Rabbah: hi Jona - welcome to the commnity https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817239027800 ---- 2019-07-22 17:41:10 UTC - Rodric Rabbah: if you’re paying the bill to operate the openwhisk deployment, you will have a fixed cost for the control/data plane… and depending on setup, a cost for the resources to run containers https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817270029000 ---- 2019-07-22 17:41:22 UTC - Jona: What I don't understand is how I can install OpenWhisk [somewhere] and not pay for idle resources https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817282029600 ---- 2019-07-22 17:42:22 UTC - Rodric Rabbah: in that sense, we might need to clarify the docs… you have to run at least an edge server and a resource manager so the costs for a self hosted instance cannot be zero https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817342031100 ---- 2019-07-22 17:43:03 UTC - Jona: Yeah I totally get the "master" node needing to opeate 24/7, but my question was more toward subsequent containers :slightly_smiling_face: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817383031900 ---- 2019-07-22 17:43:45 UTC - Jona: Is there any infrastructure/service able to instantly scale (and remove) containers under 125ms ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817425033300 ---- 2019-07-22 17:43:46 UTC - Rodric Rabbah: the current project’s capabilities fit into one of two modes: 1. deploy a fixed capacity, 2. manually add/remove capacity. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817426033400 ---- 2019-07-22 17:44:17 UTC - Rodric Rabbah: by fixed capacity i mean number of vms for running user containers, a vm may be able to run say up to 16 containers (at the same time) https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817457034000 ---- 2019-07-22 17:45:20 UTC - Jona: Sorry, I am quite new to serverless and faas in general, but how is this different than Kubernetes ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817520034800 ---- 2019-07-22 17:45:55 UTC - Jona: From what I understand, k8s do this management out of the box right ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817555035700 ---- 2019-07-22 17:46:26 UTC - Rodric Rabbah: kubernetes is a general orchestrator for containers openwhisk’s control and data plane can be deployed on top of kubernetes but openwhisk does not use kubernetes in production settings for managing the user’s function containers because it’s too slow for short running functions https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817586037000 ---- 2019-07-22 17:46:47 UTC - Rodric Rabbah: some projects like google’s knative are trying to unique kube and serverless and faas https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817607037600 ---- 2019-07-22 17:47:09 UTC - Jona: *because it’s too slow for short running functions* You got my attention. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817629038300 ---- 2019-07-22 17:47:28 UTC - Rodric Rabbah: others here who work on kantive and openwhisk can tell you more/better https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817648038700 ---- 2019-07-22 17:47:58 UTC - Jona: Would openwhisk help to provision resources faster then ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817678039100 ---- 2019-07-22 17:48:05 UTC - Rodric Rabbah: _it does_ https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817685039300 ---- 2019-07-22 17:48:28 UTC - Jona: Hmmm https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817708039600 ---- 2019-07-22 17:48:44 UTC - Michael Behrendt: yep, it bypasses the kube scheduler and replaces it with a daemonset running on each node https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817724040300 ---- 2019-07-22 17:48:47 UTC - Jona: But only on-premise then ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817727040400 ---- 2019-07-22 17:49:57 UTC - Jona: What I don't understand is, with actual state-of-the art IAAS cloud providers, openwhisk might not help due to limitations of underlying platforms. Am I correct ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817797041700 ---- 2019-07-22 17:50:47 UTC - Jona: For example: on AWS, even using openwhisk on EKS, would not help me reduce resource provision latency https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817847043100 ---- 2019-07-22 17:50:54 UTC - Rodric Rabbah: i dont understand the q there are existing production deployment of openwhisk on kube which deliver < 10ms time to provision a container for a function https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817854043300 ---- 2019-07-22 17:51:21 UTC - Rodric Rabbah: if you deploy on a VMs or mesos or some other platform, the numbers might vary https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817881044000 ---- 2019-07-22 17:51:31 UTC - Rodric Rabbah: i guess we should ask, what are you trying to do? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817891044500 ---- 2019-07-22 17:52:06 UTC - Jona: Yeah, sorry, my goal is to understand under what circumstances I can comply with the homepage claim: "don't pay for idle resources" https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817926045600 ---- 2019-07-22 17:52:36 UTC - Jona: I would like to get around AWS Lambda limitations https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817956046000 ---- 2019-07-22 17:52:54 UTC - Jona: Like the 15mn timeout https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817974046400 ---- 2019-07-22 17:53:06 UTC - Jona: But also use of GPU https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817986046700 ---- 2019-07-22 17:53:14 UTC - Jona: which is not offered so far https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563817994046900 ---- 2019-07-22 17:53:33 UTC - Jona: I am basically trying all FaaS solutions on earth https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818013047600 ---- 2019-07-22 17:53:35 UTC - Jona: ahahaha https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818015047800 ---- 2019-07-22 17:53:49 UTC - Jona: And came across openwhisk a few hours ago https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818029048200 ---- 2019-07-22 17:54:29 UTC - Jona: All FaaS are quite cool, except... I have to size my servers AND PAY FOR IDLE RESSOURCES https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818069049100 ---- 2019-07-22 17:55:07 UTC - Jona: The really good side of AWS Lambda is that it just works and scales seamlessly from 0 https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818107049900 ---- 2019-07-22 17:55:42 UTC - Jona: And this is what attracted me here: "don't pay for idle resources [as much as possible]" https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818142050800 ---- 2019-07-22 17:55:59 UTC - Jona: (GPU is really expensive) https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818159051600 ---- 2019-07-22 17:56:05 UTC - Rodric Rabbah: we should soften that claim on the website… if you’re operating your own instance, you will incur costs. openwhisk does not have a built-in VM/infra scaler in the project yet this may be useful reading <https://medium.com/openwhisk/the-serverless-contract-44329fab10fb> https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818165051900 ---- 2019-07-22 17:56:10 UTC - Michael Schmidt: If anyone has any docs on the knative vs openwhisk sutff @Rodric Rabbah is talking about that'd be cool https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818170052200?thread_ts=1563818170.052200&cid=C3TPCAQG1 ---- 2019-07-22 17:56:27 UTC - Rodric Rabbah: @Markus Thömmes might be a good reference. https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818187052500?thread_ts=1563818170.052200&cid=C3TPCAQG1 ---- 2019-07-22 17:56:32 UTC - Michael Schmidt: with the trade off of running just docker container vs the function son k8s https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818192052800 ---- 2019-07-22 17:57:36 UTC - Jona: Like I said, I get that one should pay for orchestrator's resources (like an ELB) https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818256053500 ---- 2019-07-22 17:59:19 UTC - Jona: My point is more like: what would be the time to fire a new container to compute my job ? and most importantly, how (which provider/setup) ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818359055400 ---- 2019-07-22 18:01:05 UTC - Jona: If you tell me that openwhisk can provision 100 instances based on 100 API requests in say 100ms, then execute the job... I'm in !! https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818465057200 ---- 2019-07-22 18:02:05 UTC - Jona: The minimum I could get so far was with Amazon Fargate, and it provisioned 2 instances in about 1 minute and 30 sec... https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818525058500?thread_ts=1563818525.058500&cid=C3TPCAQG1 ---- 2019-07-22 18:02:21 UTC - Rodric Rabbah: if you want an experience and capabilities as close to aws lambda, i think openwhisk is your answer. to get that kind of performance you have to provision your deployment accordingly but certainly ibm’s offering can do that @Michael Behrendt can say more https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818541058800 ---- 2019-07-22 18:03:47 UTC - Jona: Hmmm so... let's put it that way: how is IBM serving Fn ? Are they actually paying for idle resources ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818627059900 ---- 2019-07-22 18:04:40 UTC - Jona: Also, I didn't see GPU on IBM offering :sweat_smile: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818680060500 ---- 2019-07-22 18:04:52 UTC - Michael Behrendt: a recent test created a 1000 instances in 3 seconds, for example yay : Rodric Rabbah, Jona https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818692060900 ---- 2019-07-22 18:05:30 UTC - Jona: Nice. Using IBM ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818730061500 ---- 2019-07-22 18:05:37 UTC - Michael Behrendt: yep https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818737061700 ---- 2019-07-22 18:07:11 UTC - Michael Behrendt: of course, just as a data point -- depending on what you optimize for, you can even drive it further https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818831062500 ---- 2019-07-22 18:08:26 UTC - Michael Behrendt: you mean container instances or VMs? 1m30s for 2 instances sounds like VMs -- is that correct? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563818906063800?thread_ts=1563818525.058500&cid=C3TPCAQG1 ---- 2019-07-22 18:10:33 UTC - Jona: What I don't get is: say I open IBM account right now... and push my `helloworld.go` to their FaaS service. Whenever a user requests "<http://ibm.endpoint.com/my/helloworld|ibm.endpoint.com/my/helloworld>", do they actually PROVISION a new resource on-the-go ? or are they actually load balancing the user to redirect him to an already running resource ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819033066400?thread_ts=1563819033.066400&cid=C3TPCAQG1 ---- 2019-07-22 18:11:08 UTC - Michael Behrendt: re GPUs -- which use case are you shoting for (just out of curiosity) ? Many of the common use cases i'm seeing need multiple gpu vs just a fraction of it. What's your take on that? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819068067100 ---- 2019-07-22 18:11:33 UTC - Jona: Computer vision https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819093067400 ---- 2019-07-22 18:11:45 UTC - Jona: For inference https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819105067600 ---- 2019-07-22 18:12:11 UTC - Michael Behrendt: you mean RPA, for instance? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819131067800 ---- 2019-07-22 18:12:19 UTC - Michael Behrendt: or just vision in general? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819139068100 ---- 2019-07-22 18:12:53 UTC - Jona: No, just inference of computer vision models https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819173068600 ---- 2019-07-22 18:12:58 UTC - Michael Behrendt: there is no provisioning for each and every request. the system is trying to reuse container instances as much as possible, for security reasons only within the context of a give user. also, it preprovisions containers with runtimes, to have them ready to go when a request comes in https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819178068800?thread_ts=1563819033.066400&cid=C3TPCAQG1 ---- 2019-07-22 18:13:19 UTC - Jona: But rather advanced ones involving heavy computation use https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819199069300 ---- 2019-07-22 18:13:39 UTC - Jona: CPU would take 1 or 2 minutes to answer https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819219069600 ---- 2019-07-22 18:14:01 UTC - Jona: GPU takes <1s https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819241070100 ---- 2019-07-22 18:14:06 UTC - Michael Behrendt: however, when there are 1000 concurrent requests to be processed, a 1000 container instances will be running in the extreme case https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819246070200?thread_ts=1563819033.066400&cid=C3TPCAQG1 ---- 2019-07-22 18:14:28 UTC - Michael Behrendt: cool -- yep, that definitely makes sense https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819268070700 ---- 2019-07-22 18:15:09 UTC - Jona: This is why I crave to find "not pay for your idle resources" https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819309071200 ---- 2019-07-22 18:15:21 UTC - Jona: idle GPU cost like WAY too much https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819321071500 ---- 2019-07-22 18:16:20 UTC - Jona: I have yet to find a solution to provision GPU instances under 100ms https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819380072400 ---- 2019-07-22 18:16:29 UTC - Jona: I thought OpenWhisk might help :slightly_smiling_face: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819389072700 ---- 2019-07-22 18:17:02 UTC - Michael Behrendt: yep, it could be what you're looking for. Iirc, there were some folks in south korea who had implemented gpu support for openwhisk https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819422073400 ---- 2019-07-22 18:17:11 UTC - Michael Behrendt: let me see if i can still find that https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819431073700 ---- 2019-07-22 18:17:26 UTC - Jona: That would be awesome ! https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819446074000 ---- 2019-07-22 18:24:21 UTC - Michael Behrendt: i haven't exactly found what i was looking, but as part of that i stumbled across this....would this be relevant to you? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819861074800 ---- 2019-07-22 18:24:21 UTC - Michael Behrendt: <https://github.com/5g-media/incubator-openwhisk-runtime-cuda> https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819861074900 ---- 2019-07-22 18:26:57 UTC - Jona: Totally !! https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820017075200 ---- 2019-07-22 18:27:02 UTC - Jona: Thank you ! https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820022075400 ---- 2019-07-22 18:30:13 UTC - Jona: However, sorry to insist on this but: I understand openwhisk might create containers under 100ms but UNDER AN ALREADY RUNNING environment. But how can I scale RESOURCES (hardware/machines) from 0 to 1 under 100ms ? is that even feasible ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820213078700 ---- 2019-07-22 18:31:18 UTC - Jona: Or do you guys use a specific provider that bill per-container usage or something like that ? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820278079700 ---- 2019-07-22 18:32:04 UTC - Michael Behrendt: this is what i responded in a sub-thread -- does that address what you're looking for? https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820324080300 ---- 2019-07-22 18:32:05 UTC - Michael Behrendt: <https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563819178068800?thread_ts=1563819033.066400&cid=C3TPCAQG1> https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820325080400 ---- 2019-07-22 18:32:30 UTC - Michael Behrendt: you can't scale machines in less than 100ms https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820350080800 ---- 2019-07-22 18:32:45 UTC - Michael Behrendt: at least to my knowledge https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820365081100 ---- 2019-07-22 18:32:57 UTC - Michael Behrendt: there are some techniques that comes close to that https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820377081600 ---- 2019-07-22 18:33:09 UTC - Michael Behrendt: however, they in turn rely on baremetal machines being available https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820389082000 ---- 2019-07-22 18:33:17 UTC - Michael Behrendt: so you'd have to pay for them https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820397082500 ---- 2019-07-22 18:33:24 UTC - Michael Behrendt: ie you wouldn't win anything https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820404082800 ---- 2019-07-22 18:33:49 UTC - Jona: Yeah that was what I was thinking https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820429083000 ---- 2019-07-22 18:34:24 UTC - Michael Behrendt: the best way is to solve this by economies of scale and law of large numbers....but at a certain point, there is no magic https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820464083900 ---- 2019-07-22 18:35:23 UTC - Jona: But as IBM offers FaaS, I was hoping that maybe some other service could bill per GB-s usage... https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820523085000 ---- 2019-07-22 18:35:56 UTC - Jona: But hey, I can dream :slightly_smiling_face: https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820556085500 ---- 2019-07-22 18:36:27 UTC - Jona: Anyway thank you for all your replies, I'll try my way with OpenWhisk CUDA https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563820587086200 ---- 2019-07-22 21:12:45 UTC - Michael Behrendt: :slightly_smiling_face: more than happy to help. if you have any follow-up questions, pls don't hesitate to reach out. Would be great if you could share your experience with CUDA @Jona +1 : Jona https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1563829965087400 ----
