3 ms·

Thanks for the insight. How might someone in a position to spin up services for this be able to think about capturing this sort of a market? I work at a DL/ML

Thanks for the insight. How might someone in a position to spin up services for this be able to think about capturing this sort of a market? I work at a DL/ML hardware company; and many here on HN would know that packaging DL libraries correctly is a nightmare and a half. In fact a good chunk of our value prop is offering an open source pre-baked bundle. It's great (and I've deployed it on some friends' machines and they love it) but it's, let's say three-quarter-baked and moving it into a "fully supported" model with domain-specific expertise isn't something that our company has figured out how to transition into (also not easy given our company is really small), and given business strategy and software strategy are independently difficult enough problems. Do you (or any others) have any suggestions?

mooreds 7y ago
> Thanks for the insight. How might someone in a position to spin up services for this be able to think about capturing this sort of a market? Obligatory (from 2014) "Why There Will Never Be Another RedHat: The Economics Of Open Source": https://techcrunch.com/2014/02/13/please-dont-tell-me-you-want-to-be-the-next-red-hat/ https://techcrunch.com/2014/02/13/please-dont-tell-me-you-wa...
- dnautics 7y ago
  Thanks for the link!! Luckily it's not the only path for growth... As I mentioned we're a hardware company so it's more like we want to wind up as a Dell (or in a worse scenario, IBM, which isn't too shabby)
  - jabl 7y ago
    As the article the grandparent links to shows, RH is maybe the singular example of a big successful pure OSS software company. It seems almost all other big successful OSS companies have adopted some kind of 'hybrid' strategy such as open core, many after first unsuccessfully trying to emulate RH. To a hw company, I think the issue of how to monetize SW is to some extent clearer as well as pretty different than for a pure sw company. You're selling a tangible physical product, but customers are not interested in the raw hw but the complete package of hw plus supporting sw (drivers, SDK, whatever) + support. So you have to choose what is the appropriate model for you. E.g. - Use profits from HW sales to develop OSS that makes the HW more useful to customers, and thus increases HW sales. For instance, one argument in favor of this would be that drivers included in the upstream kernel and user-space software in distros makes it easier for customers to use your HW. And you'll get OSS brownie points which might also help drive sales. - Or make the SW free but not OSS, in case you're worried that your competitors could just take your OSS and use it with their HW. - Or make proprietary SW that you sell in addition to the HW. I'm not sure customers care about how you split the total bill they're paying, and free (as in beer) software is certainly a lot easier to deal with for customers (e.g. license hassles). But again, it depends. - Oh, and another potential advantage of the OSS SW model is the 'commoditize your complement' angle (do a web search on that phrase if your unfamiliar with it). tl;dr You can use OSS SW to undercut a pure SW competitor, as your income is protected by your 'HW moat'.
freedomben 7y ago
This is purely a thought at this point, but I have seen a few small companies in situations like yours be able to get some adoption by: 1. Having a product that solves a real need for the buyers (and preferably one they already know they have) 2. Getting that product "certified" with a partner, such as OpenShift. I cannot get specific, but there are a few companies I've seen that developed a clever solution to a problem, packaged it into a Kubernetes Operator, got it certified with OpenShift, and because they had that logo they got the audience with the people with purse strings. At that point it's classic salesperson strategy. A lot of big customers right now want to be able to point-click install stuff easily into their existing platforms/stacks, and then "check a box" in their list of requirements. If I were in your shoes, I would look at how people are already using it (OpenShift? AWS? Azure? Google Cloud?, etc and go from there. Getting "certified" is usually a mutually beneficial arrangement. It will take a bit of work with bureaucracies (which will annoy you greatly), but the payoff can be huge.
pabs3 7y ago
There is a team at Debian forming to package machine/deep learning software: https://salsa.debian.org/deeplearning-team https://salsa.debian.org/deeplearning-team For now the best way to contact the team is the debian-science mailing list: https://lists.debian.org/debian-science/ https://lists.debian.org/debian-science/ An example of some recent work is this update on packaging pytorch: https://lists.debian.org/debian-science/2020/04/msg00070.html https://lists.debian.org/debian-science/2020/04/msg00070.htm...

mooreds 7y ago

> Thanks for the insight. How might someone in a position to spin up services for this be able to think about capturing this sort of a market? Obligatory (from 2014) "Why There Will Never Be Another RedHat: The Economics Of Open Source": https://techcrunch.com/2014/02/13/please-dont-tell-me-you-want-to-be-the-next-red-hat/ https://techcrunch.com/2014/02/13/please-dont-tell-me-you-wa...

dnautics 7y ago
Thanks for the link!! Luckily it's not the only path for growth... As I mentioned we're a hardware company so it's more like we want to wind up as a Dell (or in a worse scenario, IBM, which isn't too shabby)
- jabl 7y ago
  As the article the grandparent links to shows, RH is maybe the singular example of a big successful pure OSS software company. It seems almost all other big successful OSS companies have adopted some kind of 'hybrid' strategy such as open core, many after first unsuccessfully trying to emulate RH. To a hw company, I think the issue of how to monetize SW is to some extent clearer as well as pretty different than for a pure sw company. You're selling a tangible physical product, but customers are not interested in the raw hw but the complete package of hw plus supporting sw (drivers, SDK, whatever) + support. So you have to choose what is the appropriate model for you. E.g. - Use profits from HW sales to develop OSS that makes the HW more useful to customers, and thus increases HW sales. For instance, one argument in favor of this would be that drivers included in the upstream kernel and user-space software in distros makes it easier for customers to use your HW. And you'll get OSS brownie points which might also help drive sales. - Or make the SW free but not OSS, in case you're worried that your competitors could just take your OSS and use it with their HW. - Or make proprietary SW that you sell in addition to the HW. I'm not sure customers care about how you split the total bill they're paying, and free (as in beer) software is certainly a lot easier to deal with for customers (e.g. license hassles). But again, it depends. - Oh, and another potential advantage of the OSS SW model is the 'commoditize your complement' angle (do a web search on that phrase if your unfamiliar with it). tl;dr You can use OSS SW to undercut a pure SW competitor, as your income is protected by your 'HW moat'.

freedomben 7y ago

This is purely a thought at this point, but I have seen a few small companies in situations like yours be able to get some adoption by: 1. Having a product that solves a real need for the buyers (and preferably one they already know they have) 2. Getting that product "certified" with a partner, such as OpenShift. I cannot get specific, but there are a few companies I've seen that developed a clever solution to a problem, packaged it into a Kubernetes Operator, got it certified with OpenShift, and because they had that logo they got the audience with the people with purse strings. At that point it's classic salesperson strategy. A lot of big customers right now want to be able to point-click install stuff easily into their existing platforms/stacks, and then "check a box" in their list of requirements. If I were in your shoes, I would look at how people are already using it (OpenShift? AWS? Azure? Google Cloud?, etc and go from there. Getting "certified" is usually a mutually beneficial arrangement. It will take a bit of work with bureaucracies (which will annoy you greatly), but the payoff can be huge.

pabs3 7y ago

There is a team at Debian forming to package machine/deep learning software: https://salsa.debian.org/deeplearning-team https://salsa.debian.org/deeplearning-team For now the best way to contact the team is the debian-science mailing list: https://lists.debian.org/debian-science/ https://lists.debian.org/debian-science/ An example of some recent work is this update on packaging pytorch: https://lists.debian.org/debian-science/2020/04/msg00070.html https://lists.debian.org/debian-science/2020/04/msg00070.htm...