New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
I feel that's a fundamental misinterpretation. Yes, software raid can benefit from write caching, as well as other performance benefits, just like hardware raid. However, A, those benefits come as a result of some forward planning (which, in the case of hardware RAID, is conducted by the manufacturer, but with software raid relies on the user and/or kernel to configure) and B come with the associated risks of unwritten data.
It seems like you said the same thing as rds100 (you can't have write caching on software RAID; you can always disable fsync, but the assumption is that the caching still provides durability, and that's only possible with BBU).
https://en.wikipedia.org/wiki/Flashcache
Quite simply, don't believe anyone about claimed redundancy and plan for the worst case.
So @jarland, if I understand you correctly here is what you are trying to say:
Cloud != redundancy. i.e., "cloud" VPS is no different than ordinary VPS (other than you can add/remove/destroy at any time), the VPS itself may or may not be redundant and if you want to achieve HA you will have to set up your application to use multiple "cloud" VPS with failover.
Is that correct?
Yes.
Don't confuse the marketing hype that Cloud = More than a VPS (kvm) on hourly billing.
Keep backups, and hope for the best.
Absolutely. That isn't to discourage providers from trying to provide high availability at the VM level but in my experience the best and most functional solutions for high availability are focused on application layers.
VM high availability is a great thing, but not as easily achieved as some would like to believe. Every time I've seen it attempted at a low price the result was that the provider created such an overly complex setup that they themselves failed to recognize that they actually added points of failure, rather than removing them.
Granted, HA at the VM level is MUCH easier when we're talking about containers instead of actual virtualization.
Distributed replicating block devices (Basically network raid1) allows you to achieve redundancy across machines as long as both machines do not have a point of failure at the exact same moment. Well that's the concept we utilized at least and has worked great so far and is affordable (Just as cheap as Software Raid 10 - You just need that additional node to replicate the failed host).
For usable performance, it is not just as cheap as software raid, since you need at least a 10Gbps network for distributed storage, but it is close to the cost, other than you need 3 nodes to make a quorum, instead of 2 sets for raid10, it is more similar to raid5
Hey if you've found the solution for low cost redundancy that works flawlessly with full virtualization and has lower failure rates than RAID controllers, by all means you'll have me sold and I'll be a customer singing your praises (I mean that genuinely). I've just lived long enough to see first hand that it doesn't always play out as well as it sounds. Put a few thousand hypervisors through it and run it for a few years, let me know if your failure rates were significantly lower than average RAID based systems.
I'd honestly love to be wrong in believing that the failure rates, in combination with problems added by the overhead of the setup, are not better at scale.
Sounds like you should invest in some cloud insurance!
Get your first mortgage today from our new Cloud Banks! High available loans which are in RAID safes(your money and online banking details are encrypted with NSA approved Caesar ciphers).
Your money is safe in with Our New Cloud Banks!
Join now and spin up your first loan today! 55 seconds to get your loan approved!
Cloud Bank! A new spin on lending.
How many copies of my $$$ do you keep? Or do I have to backup it myself?
"Cloud Bank -- Save it for a rainy day!"
or
"Cloud Bank -- Cloudy with a chance of RAID failures"
We print as much as money as we want. Please contact our director, Mr Ponzi, to have your questions answered.
Question for the DO users:
Does DO support available zone for the storage? Which means if two volumes are in the different available zones, they are guaranteed to be persisted into different physical storage?
Bu't what about waiting for 6 confirmations on the blockchain?
That's outside the scope of marketing documents. Please wait here on one leg while I return from my 5 year holiday.
In the mean time, please consult our sales director, Mr. Web Guyen, for advice on our fine affordable loans.
The title of this topic is a bit sensationalist. Why not append the word article or something along those lines? Keep LET drama here.
It might sound strange, but I spend more money on backups than on originals. I learned it the hard way, as most do, but at least I learned...
Each hypervisor has it's own RAID storage, and you generally should never have two droplets on one hypervisor (as in it isn't impossible, but the system attempts to avoid this specifically, at a best effort priority).
Thanks for the info. So basically they try to randomize the droplet deployment, but without specific SLA. And also the storage is not centralized with SAN.
That's lame.
Correct.
SLA is 99.99%.
Correct.
Sorry you feel that way, but having been the victim of many SAN failures, and witnessed often the horrible IO latency of SAN in less than optimal conditions, I personally prefer it as a customer (always be a customer of your product).
@jarland what do you do with you dos?
I still play DOS games with ScummVM.
I'd venture to say that the percentage of customers who lose data at DO is ridiculously low compared to any LET/LEB type provider. And for how much custom stuff DO has made and uses and how much more complex their systems are compared to the average WHMCS+SolusVM provider on here, I would trust them to run any production site. That said, I don't use them for that right now because I have fully managed servers for that elsewhere just so I have less work to do.
All that said, it is a bummer for OP to lose his stuff - but hey, that's what backups are for
HOLY SHIT IT IS @SHOVENOSE
DUN DUN DUN
what?
There are certainly SAN-less clouds. I heard a talk on SQL Azure and they stated they do not use a SAN on Azure. Each server has its own local SSD array and the inter-node networking is insane enough to support their "six copies of data around the world" SLAs.
Of course, Azure doesn't lose people's data...
Then again, you have to actually PAY to use Azure. No one pays to use DO, do they? I know several people who've been with them for several years without putting a penny into it and still have credit to burn. On that basis, DO is worth the money.
At DO, credit falls out of the sky. Hence the big focus on marketing the term cloud. Speaking of that, I've got about $75 there that's probably never going to get paid out in credit(pending referrals).