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
No as i noted many providers here go with new barebone servers vs used hardware. Your missing my point. Also as I showed a 16core single socket AMD EPYC 1u barebone server is 1.500 euros hardly to be considered excessive. Especially when a provider will be expanding anyway.
As i also said before encryption of data isn't something new. It isn't something unproven and it isn't considered unreasonable. Many people complained when Google chrome started calling out sites that didn't enable https but most would agree now it's worth while to have an encrypted internet space. Many encrypt their hard drives and data. Encryption isn't something mystical for us to have a debate on the merits of it. Many here say "don't trust a VM as you can dump the memory" well we now have a solution for that and it's something we trust for everything else.
The argument of "if you need to protect your data then don't put it on a VM" is the same as saying "if you have nothing to hide then who cares who reads your data". You don't need to be hosting highly sensitive data to want it secured. Additionally as I've also noted I believe in a multi tiered setup with redundancy and backups. Something's make sense on a vps something on a dedicated server and some things on a shared host. But that doesn't mean I won't care if my vps data is dumped or my shared host takes my sql database and shares it out. You want your data secured and encrypted. It's a simple concept. The missing link to the vps environment is encryption of the ram. Simple as that.
Again.. Were just talking about the ability to encrypt the servers memory for access by a VM.. Intel or AMD either case let's get it done. It just do happens though that AMDs solution has this built in and is more cost effective than Intel's.
Really? Okay there Joe.. Your the guy that yells "if you don't like it then leave the country" aren't you? The point is there is a solution out there to prevent loss of data. I said already I trust my providers and admins. I can't predict the future though, no one can. As such it makes sense to take steps to protect the data. Why put a password on your server? To prevent unauthorized access. Same reason you encrypt server VM memory!
@jsg
Thanks for taking the timer to clarify your position. Though you said it best that memory encryption needs to be put into use to see it's true merits. The are very little reasons to not test it other than concern for faults. However we're dealing with enterprise grade hardware here not consumer. A fair amount of testing and research has already gone into this. No one here that's used EPYC or online has stability concerns. If you advertise the memory encryption as a beta program with regular backups of data I'm confident many would be willing to give it a try.
@Golbinex
Fortunately AMD unlike Intel allows you to disable the PSP which in the server and enterprise space is quite useful. Many here use iDrac and IPMI for example. It's nothing knew. But i do prefer that it's ARM based, separate from the core processor and able to be disabled via bios.
@willie
Consumer vs enterprise. Again big difference. That fault was never found on EPYC additionally there was extra testing and verification on EPYC that's why it came out after ryzen desktop processors. You can't take an issue with a consumer part that was quickly resolved and RMAs allowed and compare that to an enterprise solution that never had the issue.
Not at all, there is a huge difference between those two ideas. You're expecting providers here on LET to go out and spend $4000 on a server not taking into account the ROI for brand new hardware. If clients were paying $50 a month then of course that's an option, if clients are paying us $2 per month then they have very little say in the infrastructure which comes back to my original statement, if you don't trust the provider (or the hardware) then go elsewhere. You can't sit here and say "providers should buy X hardware" when buying that hardware will put them out of business.
In an ideal world buying the latest and greatest is the route to take, but in the real world not every company does that (especially companies offering LET pricing for their services). Just because the company doesn't buy XYZ hardware doesn't mean they hate their customers.
That's only half the truth. The other half is: Nope implementing memory encryption in hardware and in the context of a multi-core processor IS quite new.
One important difference is that between encrypting payload and encrypting the memory the processor is working with. Just look at your Google/https example, even leaving aside certain ssl problems: If some crypto in https is found to not work as it should you just switch to another algorithm in your config file. If however memory encryption is found to have some problem then what? In the very luckiest of cases one would need to push a firmware update out.
I see two issues with that:
"A fair amount of testing and research..." - it is exactly that attitude that has brought us into the poor situation we are in. I bet that intel did and does a "A fair amount of testing and research" yet we ended up having Meltdown and Spectre gen 1 and 2 to name just one example.
Additionally even when looking at OS support I'm not really happy. It's not that long ago when "we support Windows X and Y" was all intel offered and even linux didn't get much attention. There is a gazillion of weird combinations out there that AMD simply can't all test or even just all know.
My second point is "not being secure and not knowing it is WORSE that not being secure but at least knowing it". You see, people do not just use assumed to be safe stuff but they bank on it. Example: process separation. Quite a few developers relied on the processor keeping things nicely separated and created safety mechanisms relying on that like for example having a daemon run as user nobody after having spun off a thread (or process, no matter) running as root and being in charge of sensitive stuff.
And btw we shouldn't forget why we want memory encryption, especially for VPSs, in the first place. To put it bluntly: some of us want a new safety bet because all the other safety bets were largely lost.
Fine lets play your game then.
A worthwhile used barebones 1u 2x 8core Intel xeon server from 2012-2014 without ram such as a E5-2690 will be roughly $500
https://www.ebay.com/itm/Variant-1U-Supermicro-Server-X9DRI-LN4F-Rev-1-20-up-to-2x-E5-2690-V2-256GB-RAM/132570042432?hash=item1eddca5040:m:mtK69MxFZxZch9DBBdjPMFg
A Barebones 1x AMD EPYC 7000 series 1socket with 16core (32threads) is about $1,500 currently.
Lets leave the fact that the EPYC processor has more PCI Lanes for NVME, supports NVME raid from the start without additional "unlock keys" like Intel and supports 2tb of ram vs 256gb on the intel.
That's a delta of about $1,000 and an additional 6 years of life (assuming you go with the first release of the Xeon 2690 from 2012).
Lets assume you go with 16GB of ram in both setups. That's low for sure but definitely feasible for anyone looking to get into the LET game. At a rate of 512mb, 1core for $3.99 a month for a KVM (those numbers sound familiar to you?). You can easily put in 25 users to generate about $100 a month.
That means it'll take about an additional 10 months to make up the difference between the new EPYC barebones server and the 2012 2x Dual XEON processor server. HOWEVER the setup is 72 months newer and more secure. I would assume most providers intend on being in business longer than 10 months.. Giving up 10 months to gain an additional 62 months on the life of the hardware should be an easy business decision.
So your using less power, gaining an additional 6 years of life on the hardware, and providing a more secure platform with a ROI difference of 10 months.
So tell me again why it would make more sense for someone to buy a 2012 2x XEON vs a 2018 AMD single socket Epyc with the same cores but with more features and a longer potential shelf life of an additional 6 years?
@jsg
But none of your arguments support why we shouldn't be testing and pushing Virtual Memory encryption forward. To put it simply your saying the old stuff had issues so why bother with new stuff? Isn't the whole point to learn from previous mistakes and move forward to new security solutions? Why fight it? Why not at least test the solution? Once again the providers here that have implemented EPYC only have praise for it. The reviewers and testers online also have nothing but praise for yet.
Yet you without any experience with the platform believe you are in a position to say it's not worth implementing because you don't trust it despite never having even tested it?
@sureiam
I see a problem. You seem again to be aggressively on some mission with only one outcome acceptable while I'm here for a open discussion. Plus you bend what I'm saying and simply ignore inconvenient points.
Btw your calculation above is also questionable. While I'm with you in clearly preferring the new Epyc over the old intel that's not how business works. For a start the motivations behind those purchases are totally different. The old 500$ box is purchased to (a) still get some money out of it and (b) to amortize quickly ~ less risk. The 1500$ box is purchased with a quite different mind set. The former has the priority of being cheap while the latter has the priority of being reliable and fast. Also note that the former has a ton of also cheap replacement parts available.
More importantly though that's not how the hosting business works. It's not like 25 * 4$ = 100$, bang amortized. No, it rather is 25 * 4$ minus colo minus traffic minus pro rata hardware cost recovery minus staff. Neglecting that the Epyc based VPSs are somewhat faster the cheap old 500$ intel boxen will be the better choice for many providers especially in the low end segment.
Yes but if the objective is to provide a reliable service taking down the hardware for even cheap hardware replacement isn't the objective. Getting it online and staying online as long as possible is the clear objective here.
Yes of course I'm well aware of the different associated costs with hosting servers. However generally they amount to rack size for CoLo (1u+), bandwidth, and power. Colo size and bandwidth are essentially independent. Old Xeon or New AMD EPYC it's the same requirement. However power would actually end up lower in the EPYC example.
Your argument that the low end market here doesn't care enough for their customers to anticipate an additional 10 month ROI on hardware is a bit incorrect as many customers here go with 1 year pre-paid services. If service is good they will renew for another 12 months.
Also I don't believe many hosts here are expanding with multiple nodes at a time. Sure some of the bigger boys are but the ones starting out are unlikely to buy 5x 1u servers from 2012.
@sureiam
I guess you are right with some providers and I'm right with some. Actually I KNOW how some low end providers play the game: They don't buy one cheap old box but 10 or more calculating that a few won't live long but can be put aside for spare parts.
Say 8 out of 10 work for 3 more years and they buy 20. And yes it IS a major factor whether a replacement processor costs 100$ or 1000$.
Also reliability is a relative term and reliability can still be good if you need to replace parts more frequently but have cheap spare parts available in stock. Keep in mind that in the low end segment an occasional short downtime of a single node isn't a major headache. If I pay say 3$ per month and get availability of 99.75% I won't complain.
It seems to me that the two of us largely agree in strongly favouring an AMD Epyc over an old intel and in being quite interested in security. We just attach different weights to certain points.
@jsg
Yes i agree were both essentially on the same conclusion just discussing more the merits of ROI.
In any case it's been determined that Microsoft Azure has started large scale deployment of EPYC. I'll be discussing with them to see if they've implemented SEV...
I was also reading this today it was posted last night:
https://www.servethehome.com/supermicro-as-1123us-tr4-server-review-1u-dual-amd-epyc/
What a beast for 1u at 64cores128threads and 4tb of ram! That's a lot of threads for 1u but ya not the cheapest initial investment either.
@sureiam
We might as well forget about this now.. AMD haven't done it well themselves.
https://www.theregister.co.uk/2018/05/25/amd_epyc_sev_vm_encryption_bypass/
I really think chip manufacturers need to base chips around a secure model with reusing old tricks from old architecture.
Only problem is that the hypervisors would need to be compatible and so would operating systems.
We are in troubling times.
Yep. And personally, if you understand how computers work, I think it should be clear that there simply cannot be any VM encryption solution that is totally safe from the host.
I'll wait to see what the fix is but it's definitely disappointing to say the least. I still think we should work towards a solution to Better protect and isolate VMs though.
Reading further in the topic it seems they needed to install a rogue hypervisor host for this security hole to work. I think we can agree that if every VM went down and a rogue hypervisor was installed and then all VM restored that it would register on someone's radar.
Doesn't seem that bad to me.
It goes from solving the problem to not as secure as advertised, and it's still a new product. It's exposing a vulnerability that wasn't supposed to be there and goes against the promises this cpu made. It means the developers weren't as great as they thought, and that's not okay if you're pushing this as the answer to vulnerabilities.
It's okay to be wrong you know, but if you dig your heels in without due process or at least a period of time long enough to qualify as "tried and true" then you're making a big mistake. It's popular around this place to call that kind of digging in of one's heels "shilling." At a certain point that's what it looks like. Maybe it's the next vulnerability or the next. This is only the first. Be more cautious with your stamp of approval, lest it lose it's value.
This is getting from bad to worse: remote "NetSpectre" attack concept is revealed...
https://www.bleepingcomputer.com/news/security/new-netspectre-attack-can-steal-cpu-secrets-via-network-connections/
https://misc0110.net/web/files/netspectre.pdf