Slashdot Mirror
Google's Plan To Kill the Corporate Network
mask.of.sanity writes "Google has revealed details on its Beyond Corp project to scrap the notion of a corporate network and move to a zero-trust model. The company perhaps unsurprisingly considers the traditional notion of perimeter defense and its respective gadgetry as a dead duck, and has moved to authenticate and authorize its 42,000 staff so they can access Google HQ from anywhere (video). Google also revealed it was perhaps the biggest Apple shop in the world, with 43,000 devices deployed and staff only allowed to use Windows with a supporting business case."
why use so many Apple computers when there's your own awesome Chromebook?
What a coincidence. Zero Trust is EXACTLY what I have in google.
Goobuntu runs on Macs just fine.
The rj45 jacks in the office are just plain old dirty connections to the Inet. We each have multiple OpenVPN connections on our localhost giving us access to different parts of the network depending on our roles. It's convenient because our workstations work identically wherever we are ( home, work, coffee shop ) and it's convenient when someone leaves because operations just invalidates the VPN certs and the former employee is cut off no matter where they physically are. A side effect is whenever your VPN credentials don't work you're left wondering is you're about to get fired and ops just jumped the gun haha.
I came to the datacenter drunk with a fake ID, don't you want to be just like me?
Both of my daughters have work issued Macs. One is in education and the other a tech company. When you look at the cost of a computer compared to the salary (and benefits) for an employee over the life of the computer, the cost of even an "expensive" computer is a small rounding error. In addition, the cost of protecting and cleaning up Windows computers is non-trivial and the cost of a data breach can be enormous.
This is not just a VPN, it is a VPN from a known, verified secure computer.
? MS Access... what a joke.
I don't read your sig. Why are you reading mine?
From a security perspective, Google is right about the notion that your internal corporate network being "safe" is dead. Between all the laptops, tablets, smartphones and very portable USB devices, there really isn't a secure perimeter on your network. Security needs to be applied at each entry point to the network, whether that is wired (internal or external doesn't matter), wireless or virtual.
The summary implied that the need for security devices goes away once you give up the idea of a perimeter, but that isn't the case at all. The form that security comes in may change, but you still need it. Authenticated users connecting via secure tunnels doesn't eliminate the risk of malware, so you still need IPS and anti-malware devices (Fidelis, FireEye, etc.) to keep your protect company assets from valid authenticated users.
If you can't trust any of the devices on your network, then you need to inspect 100% of the traffic entering the network.
Google lives in a fantasy world, where the WAN is as fast as the LAN. For me, both at home and in the workplace, you're talking about two and a half orders of magnitude difference. That's the whole reason all this cloud stuff, streaming (as opposed to download) video, etc all seems so bizarrely alien. You're talking about such a tremendous performance downgrade, that I just can't begin to really take it seriously.
I suppose the thinking is that they are planning for the future, when some day the WAN gets reasonably fast, where my home and business DSL line is replaced with fiber. Cool. Be ready, Google. But how are you going to spend those decades of waiting? Some cons are a little too long, IMHO.
But how much data do you really need to send to your home computer?
I deal with multi-terabyte datasets every day, and can work just as effectively from home as I do from the office since my data lives on the server and I never need to bring it down to my computer. I rarely even compile code on my local computer anymore since it's so much faster to do builds on the 16-core 32GB servers than on my little 4 core 8GB home computer (and even worse on the old 2core 4GB laptop).
Likewise, I don't have a Windows computer on my desk - I remote desktop to the Windows Terminal Server when I need to run a Windows app. At long as I'm not streaming video, it works just as well from home (~12mbit DSL) as it does from the office.
Interestingly, the company I work for is also like that. In our office, the "network" is just a regular consumer grade router (plus an expensive cisco AP). But we don't use VPNs (VPNs suck), all of our services are Internet accessible and protected independantly. So web-stuff is SSL + http authentication, email is IMAP, calendar is caldav. source code is ssh+git, etc. We have an internal SIP service (but that's also Internet connected).
Also, look at how large open source projects operate, Mozilla, Debian, Gentoo, GNOME, KDE, LibreOffice, etc. They're all a bit like big companies, but without a VPN, where everything is Internet accessible.
We don't use any internal application that's not web-based, does anyone else do that?
Why would Google buy Macs if they don't use OS X? They could use Linux on ANY cheaper computer they choose but bought Macs anyway.
I believe Google thinks like a lot of us: OS X for desktops, Linux for servers, a mix of iOS and Android for mobiles.
Because Apple makes good, attractive, hardware? Besides, hardware cost is inconsequential compared to the cost of a developer, whether his laptop costs $1500 or $3000 doesn't matter. Our entire development team uses Macbooks - and of 12 users, only two of them run OSX. One of them is even geeky enough to paste a Tux logo over the light-up Apple logo.
Since they deploy on Linux servers, it makes sense to develop on Linux. Write-once run-anywhere still isn't a reality - obscure platform specific bugs can still come back to bite you.
You're kidding, right? Google - home of the cloud - is going to worry about local storage limits on drone machines. And...again...drone machines - onboard video is probably 4x as fast as they need it to be for nearly all conditions. They've rolled out fiber in an entire town; I'm going to guess that they've got a pretty speedy wireless system on campus.
Apple hardware is very limited if (a) you're looking for a bargain and aren't on a corporate buying plan, or if you're a hardcore gamer, or if you are running massive analysis software, or you are locked into industry software packages which are platform locked. None of that is an issue for desk machines at Google.
I'm not, in any way an Apple fan, but pretty much none of the problems you state are of any consequence to their usage profile.
Is it just my observation, or are there way too many stupid people in the world?
I'll answer as best as I can
> Please tell us more about your setup.
We're a Java office in TX with a remote call center in OR and a handful of remote employees ( Chicago ).
> What type of work does the company and you do?
I'm the director of development, we're a j2ee web application development shop with special expertise in Oracle
> Approximately how many users work like this?
All of us ~30
> Does this company operate primarily as a standard physical office environment, or is this a distributed(work from home) startup?
A couple of my developers work from home 3 days a week and most of ops ( the network guys ) work from wherever and, apparently, whenever they want. They're pretty hot shit, published authors, speakers at LISA, etc so they're left alone most of the time.
> Where are the servers, on-site, datacenter, cloud?
We keep our staging and UAT servers on site and colo for production + another colo for failover
> Approximately how many servers?
I have no idea, I know we have some serious SAN gear for the databases. We probably have around 50 virtual servers in our testing setup and maybe 20-25 production server clusters with an average of 3 nodes each. Some physical some virtual.
> What type of applications are used, web, small applications like QB, MS Exchange or SQL systems?
Web applications, we develop/maintain some very large rewards and loyalty programs for the big banks. RDBMS is Oracle, email and IM is handled through Zimbra, project management is handled with Atlassen Jira self hosted.
> What are the negative aspects of this system?
The only problem i've ever faced is the VPN endpoints not staying connected. VPN connectivity becomes mission critical because without it no work can get done. I don't know what they're using for the VPN server, I know ops is a big fan of OpenBSD so it wouldn't surprise me if that's what they are using.
I came to the datacenter drunk with a fake ID, don't you want to be just like me?
No, a VPN still depends on a perimeter defense; the VPN is an tunnel through the perimeter and once the tunnel is set up, you have full access.
They buy Apples to save money?
Cue the frothing idiot tax minions....
What they're saying is that the idea of border security is a bad model. One compromised system on the inside and you're pretty much done. IDS and DPI are good ideas but they aren't effective enough. Breaking in to any corporate network is as easy as spamming it's users with social-engineering-laden email. Get them to click on a link and you own their soft, squishy, zero-day-vulnerable desktops. Keylog and steal their credentials and you've got a jumping off point to worm in to the rest of their network. It's that easy.
What they're saying is once you move to a trust-nothing model.. Why bother investing in a huge corp network when you can't trust it anyway? When you don't have big corp network what's, the advantages of running your own services over purchasing them from someone else? Like Google?
Our entire development team uses Macbooks - and of 12 users, only two of them run OSX. One of them is even geeky enough to paste a Tux logo over the light-up Apple logo.
The last time I visited Google HQ (about 5 years ago) the most common setup I saw was Thinkpads running Linux with Macbooks running Linux in a close second.
Wrong! Dogs are dumb, just easier to control and teach "tricks", since they are pack animals. Anyway, the cat's brain got twice the count of neurons than the dog's brain got.
Because we're dealing with zero trust.
That ALSO means I don't necessarily trust a 3rd party host either.
Chas - The one, the only.
THANK GOD!!!
But it's not a bad idea, it's just dumb to rely solely on it.
I can just imagine the military "Fuck the perimeter, if the enemy gets inside the base it's going to be all knives and hand to hand combat anyway. Sell the guns boys, we're all getting HUGE KNIVES!"
Mod me down, my New Earth Global Warmingist friends!
"Why bother investing in a huge corp network when you can't trust it anyway?"
Redundency in security.
And its in-hand. You can fix it, expand it, modernize it, control it, instead of shifting all that responsibility to some third party to which you are merely another customer.
Trusting nothing, protection at machine level, the user level, the application level and the data level will not do away with the corporate networks.
If anything, it may have the opposite effect, and encourage more use of such wholly-owned networks, perhaps melded with some cloud services.
But as sooner we move away from the Maginot Line mentality for our networks the better.
It may seem counter intuitive in the physical world, but a point defense system is easier to implement in computer networks than in the real world. Each computer should protect itself. Build this in from the beginning and it just happens naturally each computer, each file, each application. Because relying on the stockade to keep out the attackers hasn't actually worked that well in the physical world, and costs a boatload of money and expertise in the network world.
What good is ipv6 if we all have to hide behind firewalls forever.?
Sig Battery depleted. Reverting to safe mode.
I can just imagine the military "Fuck the perimeter, if the enemy gets inside the base it's going to be all knives and hand to hand combat anyway. Sell the guns boys, we're all getting HUGE KNIVES!"
RL military analogies often map poorly to network security space yet it rarely prevents people from making them anyway.
As the senior admin for such an outsourced network, I can tell you what will happen about 2 to 3 years after you migrate to an outsourced service like this.
"We're deprecating the ODBC connection as of January 1... no worries we've got a great new API and it accepts SQL!"
"To reduce system load and improve overall performance of your system we're limiting SQL requests to 100k rows"
"To enhance SLQ efficiency we've written our own proprietary query language called FU-SQL it's fantastic"
"We're aware that some of our customers are not happy with speed of FU-SQL so we've limited the number of joins you can make in a select statement to 1"
"To reduce costs for our customers we now bill our FU-SQL module separately, if you don't use it you don't have to pay for it! If you would like the unneeded additional FU-SQL feature it will bill for $150k/year"
"due to lack of interest FU-SQL has been discontinued, if you need mass access to your data please contact our professional service"
At this point they start doubling the price of their service every time you sign a new contract. Then your boss will ask you why your quote for migrating the network somewhere else was "A Metric Shitton of money"
Have fun with your outsourced network!
So does that mean (from your link) that men are 21% smarter than women? And women just appear smarter because they're pack animals?
I don't think you can compare it to a physical situation.
If you had secure operating systems, and encrypted data flows, and weren't listening on a bazillion ports, it would be just as easy to secure the network by securing individual computers as it would to secure the perimeter.
The problem is security is a bolted on afterthought for some operating systems (Windows), printers, storage devices, and software applications.
If we could get past that, we could stop building walls.
Sig Battery depleted. Reverting to safe mode.
I'm no expert in the field, but my understanding is that there are several models of network security based on real-world notions of security.
VPN is a part of your traditional wall security, where your typical authentication and authorization happens at each level of security zone. Once you're in, you can do anything the zone permits you to do. VPN is, as stated by others, placed at the perimeter.
BTW, full internal company-wide encryption just means putting the secure zones under a roof so no one flying overhead can see what's going on from above (e.g. big brother).
Another model of security relies on negative feedback. There are no locks anywhere, and no one has keys, but missteps have consequences. That's the security model most modern governments employ against their citizens. The levels of surveillance, strictness of the deeds, and harshness of the punishment determine the repressiveness of the model. The level of security is proportional to the amount of monitoring (a place like prison being maximum security).
There are other models, I'm certain, but like I said, I'm no expert. These are the two more prevalent ones out there right now.
Zero trust is completely different. It's almost like a double-blind experiment. There's no trust anywhere. Not the users/developers, not the administrators, not the auditors, not anyone. Authentication is fundamentally a trust-building mechanism, and a zero-trust model means authentication is obsolete (remember, encryption is merely erecting a roof over everything). Anyone can get in and do all the same things. The only difference is in the domain knowledge of the actors, which differenciates those able to do more things from less things if anything at all.
A rather dirty analogy of zero trust would be hosting an open project on Github. Anyone can go in and make modifications, but only those who know the code could make modifications that do meaningful work. And then, of the people building the code and running it, only those who who possess the ability to verify the modifications would know that they're not harmful specifically for their use cases.
Another analogy of zero trust would be to have an open e-mail account. There's no guarantees the sender is represented by the name. Every e-mail is assumed to have been read by anyone capable of entering the system. (Changing or deleting e-mails can be universally prohibited.) Such an account would be mostly useful for communications of metadata information, i.e. where and when to meet, and trivial matters.
I don't think Google's gone quite that far with their security model. They may have gotten rid of the VPN (or not...), but there are still SSH keys used for authentication and authorization, and users still need to log in to their machine to use it. After all, zero trust implies that even we the ultimate end users can't trust what's coming out of Google to be accurate (assuming that we could before--that's another debate for another time). And I don't think Google wants to make that impression.
It may be that they started with a zero-trust model, and identified the areas where trust is unnecessary, which they left insecure. At the same time, they also identified where trust is absolutely necessary, as well as the level of trust that's appropriate, and put up the necessary strength of walls to secure them, as well as levels of monitoring to see who's entering different zones. That sounds far more reasonable to me, especially considering the amount of trade and other secrets Google is holding onto.
"If a nation expects to be ignorant and free in a state of civilization, it expects what never was and never will be."