A way for startups to build a solid IT infrastructure (2017) | Hacker News Reader

A way for startups to build a solid IT infrastructure (2017)

about.gitlab.com

88 pointsPascLeRasc39 comments

If you're not paying in cash, you're paying in time and attention it takes to manage everything yourself. I would never recommend that a start-up self host email or very much else. This is a list of things to distract your team from focusing on building product and selling it.

It’s a question of risk and cash flow. Internal IT infrastructure can represent far lower burn rates which may be an excellent tradeoff or a poor one based on several factors.

Are you spending your personal savings or a 100 million dollar funding round, etc.

If $5/(user * month) for google suite is imposing cash flow problems then you should fix that rather than spend your time administering your own email server.

FWIW (this is merely a single example): my current company (which includes two ex googlers among the five founders) specifically uses no google infrastructure (no gmail, no google docs, no cloud, photos, hangout etc).

Actually I exaggerate: three of us use google search.

Edit: Someone asked why but their comment was downvoted dead so I am amending my comment.

There were three reasons; privacy, features, and dependability.

The privacy one is well discussed elsewhere. Apart from paranoia (won't get into that), there are UX issues. For example on the shared google docs for our local Boy Scout troop I've seen patient's files accidentally shared by physicians, internal corporate documents from some Fortune 50 companies and the like. Clearly nobody shared those on purpose!

For features: we certainly don't want to build a dependency on chrome (the new IE). One of the team had to move a google acquisition from AWS to gcp and swore never to use the latter again. Gmail doesn't follow standards (e.g. IMAP, adding AMP to mail)...the reasons go on. many products simply aren't best in class (e.g. zoom works much better for us than hangouts); others are (google docs / sheets may not be best-in-features but seem to have the best simultaneous editing, better than Apple or Nextcloud/Collabra when the network is flakey)

For dependability: one is of course products and features getting randomly cancelled. The other is cases where one account gets cancelled and it wipes out all connected features. These stories are scary; some have appeared on HN.

All in all it's not worth the risks.

Thanks for your insights. I used google suite only as an example, there are a great variety of other services available for email, etc, some with much better privacy features. My main point was that maintaining your own infrastructure is much more expensive than using off-the-shelf services, so the price argument further up the thread isn't really justified.

Email is unlikely to be the only service you need. Add github and your at 14$ a month per person. Keep going and it really does add up fairly quickly.

Yeah, I kind of assumed that the answer would be to solve your problems with office 365 or something similar and then move onto actually producing yourself.

I only skimmed the article, so perhaps I missed it, but it seems there is no mention of any central user authentication/authorisation solution or personal certificate storage like Active Directory. Every IT infrastructure that is expected to scale beyond few users has to include a component like this to manage users, their certificates(or keys), access rights etc.

And they're missing a ticketing system, a CRM, a wiki, an inventory database, etc, to say nothing of the actual computing infrastructure.

It's kind of insane to run a business on free software. If your CalDAV server suddenly has a bug with a partner who uses Apple/Windows/etc calendar, what are you gonna do? Drop everything else to start debugging? Call and tell them to switch calendars for you? Abandon them?

Attitudes like this are what leads to systematic vendor lock-in and monopolies. If an open source system can't communicate with a proprietary system, the bug is one of these systems not following open standards.

I get your stance - but this is about a company trying to make money. If anything(such as avoiding vendor lock-in) gets in the way of you making money, then your venture is failing.

Card carrying FSF member here, but business is business.

I don't know, not for me. I still think it's possible to think of your business also as a political statement. Sure, you have to be prepared to 'pay' for your conviction, and possibly fail because of it. Which does not seem to make sense or be acceptable to a lot of folk. It is for me though.

No one's ever been fired for buying Microsoft/Intel/Cisco. This approach looks cool, but it also seems like a lot to ask from the average sysadmin.

Regardless of what you paid for a piece of software, if it is critical to your business operations you should be paying someone to maintain it for you.

If you were a small lawn mowing company and happened to receive a free mower for your operations, you wouldn't just run it into the ground without regularly paying someone to keep it in good repair (though this analogy does raise some interesting thoughts about companies in general and their maintenance policies).

In exactly the same way, you shouldn't be using software of any kind, paid or not, without having some plan for maintenance and repair.

If free software can cover the same business needs as paid software, there's no reason to not use it. But you absolutely cannot sacrifice maintenance just because you did or didn't pay money for the software itself.

If it's cheaper to buy a new mower every year and sell the used one compared with paying someone a salary to do maintenance then which would you choose?

“If it’s a core business function — do it yourself, no matter what.”

— Joel Spolsky, In Defense of Not-Invented-Here Syndrome: https://www.joelonsoftware.com/2001/10/14/in-defense-of-not-...

Implementing your own email/file sync/identity server sounds like great advice for any startup to get going.

In your in-house programming language, of course.

I believe you are being insincere and/or sarcastic. This does not help, and it would have been more useful to make actual arguments.

It is an actual argument, he's saying that your application of this phrase makes no sense here. Are you suggesting startups write their own email server?!?!

If they’re offering e-mail as a service to customers, and trying to differentiate by the features of that service, then probably yes. Otherwise, probably not. Read the linked post to get a more nuanced view of what Joel considers being a “core business function”.

They're not, no one is talking about that except you. You're completely off topic.

This is about startups hosting their own internal IT infrastructure.

It’s an analogy.

It literally isn't, as you've presented it.

> [...], we use a private mail server.

I will always recommend against running your own mail infrastructure. The administrative expense is to high in the long run. Unless you want to employ a 24/7 on-call admin team, Office365 is always the better choice. Email encryption can still be done client side (SMIME/PGP).

"running your own mail infrastructure" - I once got responsibility for a complex email nightmare dumped on me (tens of thousands of users, hundreds of domains, hundreds of servers) and I had no idea the vast number of entertaining ways email infrastructure can be screwed up.

Mind you - I guess at a small scale it might be OK - but infrastructure like email is both vital and utterly thankless. Let someone else do it.....

Still have nightmares from my first job in help desk, sweating bullets while handling decades of Outlook .pst files for a C-level in my company. Only place they existed were on the laptop itself, and a cheap 250GB external drive.

Email sucks.

I still remember the conversation with a senior IT support guy who had been sent my way by the CIO - "<boss> says to speak to you as email is broken". "How much of it is broken?", "All of it"....

The cheap VM running sendmail with 12,000 entries in its alias file was being DOSed....

That was my introduction to crappy large email systems!

I think there should be a question of if the C-level needs or even has made recent searches on any of that email going back past 7 years.

I asked them that directly (it was a tiny company and I was like 20) and the answer was no, but that they wanted to hold onto them in the event they needed to look up old contract info.

Looking back, I should have suggested backing it up to network storage. I imagine it must have been stressful traveling or going through TSA and worrying about losing 20 years of email data.

I agree that one should not host their own mail server these days, however, do the business, clients, and employees a favor and evaluate something in other than Outlook and Exchange based email systems.

What's wrong with Office 365?

I've been running email servers for my personal and business email for nigh on two decades now. It's "an hour or two of maintenance a quarter", on average, and is nowhere near as big of a deal as many would make it out to be. Is it 100% perfect? Certainly not. Would it be 'easier' (FSVO) to just give my email to gmail/O365/fastmail? Probably. However, I feel "the net" was meant to be (and works best as) a series of decentralized systems. Feeding my data into a 'just this side of a walled-garden' ecosystem does not help the net, so I don't do it.

I agree, there are tons of good options out there, such as ProtonMail or others. As a sysadmin, who no longer works in the MSP world, managing Exchange was such a huge time-sink. There is reason basically every MSP who is worth their salt will migrate their clients to O365 or some other solution. It works. There is relatively little down time. And you barely have to manage it. It's a win-win for everyone.

Just buy a couple Synology's, that will get you 90% there.

You're being downvoted, but for small businesses Synology is pretty good. They now even come with Docker, full virtualization and Active Directory built-in, and they've had clustering and easy automated failover for years now. You pay a bit of a premium, but it does make it easy.

I just heard of this because of your comment, these little servers seem like a good deal for not having to do anything yourself really, especially for under 2000USD.

I'm with the majority of the comments here. It seems an expensive way to run your business. Initial cost is usually a small part of the overall TCO. This post was written in 2017, any updates on their experience with this infrastructure? Prove us doubters wrong.

It feels a bit like it misunderstands what IT infrastructure is all about. Identity and Access Management is noticeably missing AND core to IT infrastructure.

I'm very surprised to see that they chose Arch Linux for their base OS, very curious to see if that ever lead to issues over the last two years.

Reply on news.ycombinator.com