Hacker News new | ask | show | jobs
by a11595 2707 days ago
they don't build it this way because cloud is cheap and slow. I have a UCS farm with 2000VMs, connected to 3PB of usable all-flash across 2 arrays, replicated active/active to another array, second-hopped to cloud. That's one site. About 100k users across 2 domains - 1k? I said enterprise. A SAN is not scalable but internal disk is? You need to look up what a SAN is.

The reason there is a standing joke with your coworkers is because you don't work on important things that run the world. When 1 minute of downtime costs you over $1mil, your "ripped off" Oracle costs, your SAN costs, etc, are lost in the rounding errors. I can have a SAN have hundreds of arrays, across multiple datacenters, and dynamically grow and shrink my storage needs. It is the definition of scalable. I can have one VM farm vmotion to another VM farm 30 miles away transparently, which will sit on a different array attached to the SAN. What happens when the storage needs of your server outgrow the drives you can shove in there? There are servers and databases a petabyte in size, pushing a million IOPS. They're in charge of money. If there is corruption on the SAN, well it infrequently happens, just like with servers and memory. Twice in my vast experience. You can roll back all writes on the replication software, and usually keep an undo journal a few days long, and something like hourly snapshots. This also protects against cryptoviruses and other types of corruption.

This is why people like you work at small companies, fiddling with your cute little projects, while the world moves forward with you on the sidelines. I've been doing this for 20+ years, and have been at most fortune100 companies, in 80 countries. But yeah, your opinion, while being dismissive, is cute.

The CIO is the guy in charge of getting this stuff at enterprises. It's clear you don't understand the impact of design. I am guessing you are not at an architect level. There's a reason for that, and a reason other people - the ones you put down in your post, are making the big decisions. People like you would cost the company millions of dollars in loss per year.

1 comments

I said tenants, not users. Tiny little detail that you overlooked there. =) Last company I worked for had a €150.000.000 tech budget and about ~90k employees.

Again, if you read what I said you'll notice that I pointed out that you actually can build decent architectures using enterprisey stuff, it will however never scale horizontally in a resonable way. What happens when you have filled all your shelves in the array box? Ah, you'll need a new $250K box...

Mentioning vmotion and all... yeah so... VSAN et al... brrr... I'll put my trust in the open source world rather than shoddy software from 3rd party vendors trying to appease to the latest fad. Hyperconverge my *ss! =)

KISS is what rules but vendors are busy feeding channels and partners with impossible to penetrate acronyms leaving you in a mess sooner or later.

Focus from these vendors is to keep integration at a minimum, which means API's are usually crappy and achieving a reasonable level of automation is often a chore.

Now there is probably a little percentage of "enterprises" that uses tech in a sane way, but my bet is that the CIO you mention is a blockchain expert, as well as putting AI at the top of strategic actions to "implement" this year.

He probably have commissioned a pre-study from Accenture that outlines these strategic imperatives.

This is of course a little rant, but also true for the 90% of the 90% mentioned.

Don't be defensive and close minded. That's what's leaving most enterprises in the dust. There's just a lot of inertia inherent within certain businesses that will let them continue to burn through cash on pointless tech for yet a while.

again, cute. $250K. very cute. Try $5mil for a box, at least, after a 50% vendor discount. VSAN has nothing to do with SAN, and no one on a SAN uses VSAN. You are again showing your lack of enterprise experience, yet you are strongly trashing what you don't understand. A server sees many arrays, over a SAN. That is called horizontal scaling. You can't do that with internal disk.
Oh, 250 was for the box, no disks. 75% discount. :)

Keep it old-school!

Vsan btw is vmwares attempt to actually achieve _horizontal_ scalability.

I just don’t trust vmware with anything except for that bytecode vm. It used to rock, but time went ahead, and even as they IPO’d I thought they would end up dead in the water eventually. I had just been introduced to the wonderful zones in Solaris and it just made hypervisor vm’s seem silly.

And here we are with cgroups and friends...

Have a look at ceph if you’re interested: http://docs.ceph.com/docs/master/architecture/

Anyway, I’m out.

If your horizon extends to what vendors are prepared to sell you, take that 50 discount and run with it!

One final thought: is that disk not ”internal” to the array? This is what makes blockstorage notoriously difficult to scale horizontally. You’ll need very clever software!

GL&HF

yeah, I know what VSAN is. It's you who does not if you think you run VSAN on top of a SAN-connected cluster. Yes, the disk is internal to the array. A server can see a hundred arrays on the same HBAs. It doesn't care what array the storage comes from. I am positive at this point you know nothing about what a SAN is.

What "box" - the 42U Rack? I don't think so. The rack is always free. You then have DAs connected to the disk, and FAs connected to the SAN, which are on a pair of directors. You literally cannot get those w/o disk.

You don't know what a SAN is, you've never seen an itemized quote for an array. Thanks for your link. It's like sending a hooked on phonics link to an English professor. Cute. Keep it cute! People like you are the reason people like me get paid a lot.

For free! Good one. Cause’ that’s whats really happening... right? With a straight face?

I was certain I had nothing more for you, but this is too much fun!

You clearly don’t grasp the difference between vertical and horizontal scalability which means you have never been subjected to a bunch of scenarios requiring the latter.

Dinosaurs taking the p*ss are the reason many enterprises opt to off-shore and out-source.

yes idiot. I have worked for several vendors and sold this stuff. I have also been on the customer side buying this stuff. You're paying 250k for an empty 42U that you call "box" - you literally are lying.

there are, literally, zero enterprises off-shoring their hard IO hitting datacenters. In fact, having been in 80 different countries for these enterprises, they usually have many datacenters all over the world.

VSAN is for hyperconverged systems only. racks with a thousand 1u nodes that all have disk, connected to a fat ethernet backplane. Things on a SAN are not hyperconverged - they are on a SAN. VSAN is for tier 2 stuff that goes on hyperconverged - like web servers and DMZ things. The fastest processing is done on solid databases like Oracle or UDB on clusters of large servers, connected to a SAN. VSAN is even positioned by sales for tier 2 from all major vendors.

You literally picked up some technical words you heard around the office, googled a few things, and now consider yourself and expert so you give authoritative opinion on here about things you have never worked with. I bet you are deskside support or a code monkey, and have never architected a solution. When someone gives you a budget of 20mil and says you can average 1min of application unavailability per year or it impacts billions in company's bottom line and your whole team gets fired, do let me know. I'm sure your suggestion would be to cluster together a bunch of old dell laptops over wifi.