Looking for virtualization advice
-
@jaredbusch said in Looking for virtualization advice:
@kooler said in Looking for virtualization advice:
@jaredbusch said in Looking for virtualization advice:
Side note, how do you calculate cores for Server 2016 on a Scale system. Is it per node?
because I specifically spec'd that Xbyte system with 2x 8core procs because of Windows licensing
It's per-core from Microsoft. Scale can't do anything to change MSFT licensing policy.
Right, I meant how do you count which cores apply.
You control where things run, if you want to at least. So you only count the cores where you allow Windows to run.
-
@tim_g said in Looking for virtualization advice:
How this is done best is decided via the expertise of IT, not stakeholders.
Tim is correct and this can't be overstated. Once you run the math, you have the answer. If a stakeholder votes against the math.... that's sabotaging the company. If you have stakeholders doing that, you have bigger problems.
-
@dustinb3403 said in Looking for virtualization advice:
@GaryP have you performed a Business Impact Analysis?
That would help you to determine what solutions are the only reasonable solutions (in terms of financial spend).
If he lacks the info that he needs, then definitely. Speaking to others to determine the components for the math, yes makes total sense. Once you have those numbers and run them, then opinions should end and it's literally all math.
-
@garyp said in Looking for virtualization advice:
Yes, it is usually money but also redundancy/DR are a consideration and who is going to support it.
Either redundancy / DR is a money decision, or it's just emotional randomness that undermines the business. DR, performance, all that stuff comes down to profits and making money.
-
@scottalanmiller said in Looking for virtualization advice:
I'm a bit partial to Restoronix here, for obvious reasons.
I'm not sure I'd say that your bias for Restoronix is obvious. While after some digging around on the site I'm sure someone could find the reason, but to ensure transparency, this is a case where I'd say a flat out disclaimer of your role with Restoronix should be required each and every time you make the recommendation. Many others on the site (if not most) carry the "Vendor" tag when they participate to highlight this where in this case, you do not.
Btw, Restoronix may be the perfect solution, I'm not saying it isn't. But representing your product you may want to consider avoiding even the appearance of conflict.
For the OP who joined only a few days ago: https://www.mangolassi.it/post/324872
-
@garyp said in Looking for virtualization advice:
We are looking to virtualize the servers in the data center in our office. Since most of our applications run in two co-lo data centers, managed by a service provider, the data center in our office could almost be considered a remote office.
Currently we have mostly physical Windows servers (yes, I know it is 2017) in this data center consisting of:
Two file servers for a total of 7 TB
Security system server
HVAC system server
Warehouse conveyor control server
We need to add about 6 Linux servers for SIP and a telecom specific application.Ideally, we can move VMs to another host if one fails, but it doesn’t have to be instantaneous.
Current storage need is ~8 TB with room to grow to ~15 TB in the next 3-4 years. DAS, NAS, etc., whatever might work best for this situation.
Dell suggested a 3-2-1 architecture for over $100,000, which does not seem at all practical.
Another vendor suggested Dell’s VTRX with 3 M630 blades plus the storage.
We have started looking at HCI solutions, including Scale. StarWind and HPE SimpliVity as we do not the expertise in managing a hypervisor nor the time to manage it. Too many other projects and distractions.We will need some sort of DR solution. We will want an on-site backup and a copy going off-site to Azure, AWS or similar.
Suggestions?I've seen several business with similar requirement.
My suggestion for your scenario:
- KVM should be your hypervisor of choice. It has NO limits whatsover on anything. XS is a great choice but with much less features.
VMware essential just if you use solutions that are certified only with that platform. - Go with a just ONE Dell R740 with two 16-cores cpu, 576Gb of ram and 9x2Tb SAS ssd in raid 5. Prosupport 4h is fundamental in this configuration. You can have a machine like that for 20k$ or less (Dell just quote a pair of that for me)
- Buy a second, basic server and fill it with large spindles (raid10). Install Linux on it and made it a perfect backup target.
- Use an agent-based backup software and upload anythig to s3.
I don't thing you need any sort of HA technology. You'll thank me later for the removed complexity.
- KVM should be your hypervisor of choice. It has NO limits whatsover on anything. XS is a great choice but with much less features.
-
@francesco-provino said in Looking for virtualization advice:
- KVM should be your hypervisor of choice. It has NO limits whatsover on anything. XS is a great choice but with much less features.
VMware essential just if you use solutions that are certified only with that platform.
KVM or Hyper-V are the only things to use IMO.
- Go with a just ONE Dell R740 with two 16-cores cpu, 576Gb of ram and 9x2Tb SAS ssd in raid 5. Prosupport 4h is fundamental in this configuration. You can have a machine like that for 20k$ or less (Dell just quote a pair of that for me)
This is just stupid expensive. Almost as bad as @scottalanmiller's recommendation for a Scale cluster. In fact I would by a $30k Scale cluster over something like this.
The only information we have form the OP is 2 file servers with a bunch of data. File Servers. This does not mean SSD, nor 576GB of RAM. The OP only needs 8 7.2K NL SAS drives in a RAID 10 to achieve 16TB of space. With a H7XX controller with 1GB or 2GB of cache, he will almost never see a performance constraint caused by the RAID array. Also, they OP is obviously a Windows shop based on the "add some Linux Servers" phrase. That means anything more than dual 8 core procs will immediately also call for an increase in Windows Server licensing. You suggestion means doubling their Windows server licensing.
- Buy a second, basic server and fill it with large spindles (raid10). Install Linux on it and made it a perfect backup target.
This is way to generic. A large part of this will depend on the hypervisor and also on retention needs.
- Use an agent-based backup software and upload anythig to s3.
Likely the backup software can connect to this.
I don't thing you need any sort of HA technology. You'll thank me later for the removed complexity.
But for your proposed cost he is better off with a Scale cluster. The only reason I do not recommend it is because the hardware costs should not be anywhere close to a $30k Scale cluster.
- KVM should be your hypervisor of choice. It has NO limits whatsover on anything. XS is a great choice but with much less features.
-
@jaredbusch said in Looking for virtualization advice:
@francesco-provino said in Looking for virtualization advice:
- KVM should be your hypervisor of choice. It has NO limits whatsover on anything. XS is a great choice but with much less features.
VMware essential just if you use solutions that are certified only with that platform.
KVM or Hyper-V are the only things to use IMO.
- Go with a just ONE Dell R740 with two 16-cores cpu, 576Gb of ram and 9x2Tb SAS ssd in raid 5. Prosupport 4h is fundamental in this configuration. You can have a machine like that for 20k$ or less (Dell just quote a pair of that for me)
This is just stupid expensive. Almost as bad as @scottalanmiller's recommendation for a Scale cluster. In fact I would by a $30k Scale cluster over something like this.
The only information we have form the OP is 2 file servers with a bunch of data. File Servers. This does not mean SSD, nor 576GB of RAM. The OP only needs 8 7.2K NL SAS drives in a RAID 10 to achieve 16TB of space. With a H7XX controller with 1GB or 2GB of cache, he will almost never see a performance constraint caused by the RAID array. Also, they OP is obviously a Windows shop based on the "add some Linux Servers" phrase. That means anything more than dual 8 core procs will immediately also call for an increase in Windows Server licensing. You suggestion means doubling their Windows server licensing.
- Buy a second, basic server and fill it with large spindles (raid10). Install Linux on it and made it a perfect backup target.
This is way to generic. A large part of this will depend on the hypervisor and also on retention needs.
- Use an agent-based backup software and upload anythig to s3.
Likely the backup software can connect to this.
I don't thing you need any sort of HA technology. You'll thank me later for the removed complexity.
But for your proposed cost he is better off with a Scale cluster. The only reason I do not recommend it is because the hardware costs should not be anywhere close to a $30k Scale cluster.
You are right, I took this configuration example because the vendor just quote it. Using the same single-server pattern, you can easily build a config that fit his needs (now I read them well) with 5-6k. The generic storage server as a backup target is just an advice against pre-made NAS.
- KVM should be your hypervisor of choice. It has NO limits whatsover on anything. XS is a great choice but with much less features.
-
@scottalanmiller said in Looking for virtualization advice:
@garyp said in Looking for virtualization advice:
Dell suggested a 3-2-1 architecture for over $100,000, which does not seem at all practical.
Suggestions?
Most important suggestion, don't talk to Dell (or any vendor) as their advice is solely to mislead you. You are correct, a 3-2-1 isn't just impractical here, it's insane. It bleeds the coffers without bring any advantages, literally none, over having just a single server. In fact, it is six points of failure when only one is needed. It provides no high availability, yet makes you pay for a crazy amount of redundancy. You spend $100K to less than $10K of system.
More important, give me the names of the Dell Reps, the quote number, and the partner who recommended VRTX. They shouldn't be leading with non-HCI solutions in this space. I've got a
Re-educationTraining coming up for some Dell reps and partners and I would REALLY like some feedback on what's causing them to recomend non-HCI solutions.I can offer bribes of dinner and drinks at most major conferences where I show my head
-
@jaredbusch said in Looking for virtualization advice:
@scottalanmiller said in Looking for virtualization advice:
@garyp said in Looking for virtualization advice:
We have started looking at HCI solutions, including Scale. StarWind and HPE SimpliVity as we do not the expertise in managing a hypervisor nor the time to manage it.
That's the appropriate short list. Of those, @Scale is the one that is going to offload the most from your plate. Starwind provides HC but you are still managing the hypervisor on your own, separately. It's architecturally all together, but the management console is not.
He doesn't need hyperconvergence. Don't sell him something he does not need. There is no way to intelligently get 15TB of storage on a Scale box for a reasonable price compared to local storage.
Other HCI platforms have no problem getting 15TB in a box (I had a serious discussion with an SE about a 300TB+ per host vSAN design and all the considerations around it). I think someone was taking socket based licensing to a ridiculous logical conclusion but with 100Gbps networking becoming affordable this stuff isn't that far off.
I regularly see 90TB configurations. Starwind and other mirror based SDS systems with direct connect can scale pretty deep.
The other thing to note is if most of this data is ice cold, it may be better to put it into an archive system. CloudArray or Choehsity virutal appliances and other systems like it allow you to dump the cold data into an ingestion point (NAS share, or iSCSI) where it is dedupe and compressed and then cold data is pushed out to the cloud and tiered to an object store of your choice. This way you can run HCI with Asymmetric storage growth. For engineering shops who have to retain 10 years of stuff this is a good way to make it efficient, archive it, tell it to mirror to two different AZ's in glacier and then ignore it. By doing data reduction on the ingestion you can cut down on your cloud storage quite a bit.
He can likely get a decent solution for a lot closer to 50K than 100K. (Go single socket, and use the HCI acceleration kits).
While a single server solution is fine, being able to do non-disruptive maintenance is really damn nice, especially with systems like HVAC and security that may have compliance or safety requirements on staying online.
For telecom, pay attention to what hypervisor your platform supports. A lot are picky because of timing concerns, or how they do clocking for trans-coding. If your not trans-coding it's not normally a huge deal but trans-coding is where stuff can get weird.
Curious why DR to Azure or AWS. While there are solutions that can do it, there are cheaper/better IaaS players who offer DRaaS (Look at Veeam's partner network, there's a lot of good players there).
If your hell bent on DR to Azure and AWS, choose Hyper-V for Azure, and ESXi for AWS (as VMware on AWS will be coming out of beta soon) would be my picks. Ideally though if your doing DR to either you'd be doing it at app/PaaS abstraction layer using something like Pivitol but you are too small for this.
-
@jaredbusch said in Looking for virtualization advice:
This is way to generic. A large part of this will depend on the hypervisor and also on retention needs.
Use an agent-based backup software and upload anythig to s3.
Also will the backup software restore things like Active Directory, GPO's, SQL Schema's. Does it allow you to search for files (Good luck finding that PDF if you don't have an index!). Does it let you restore pieces of Mail from exchange etc.
Also you say DR to Azure or AWS. Copying 16TB of data there isn't DR. That BC. IF you need to re-hydrate 16TB of data at a file level over the WAN from S3 (which isn't fast to re-hydrate cold data over the WAN) it can get... fun....
-
@jaredbusch said in Looking for virtualization advice:
I don't thing you need any sort of HA technology. You'll thank me later for the removed complexity.
Hypervisor HA (at least on ESXi) isn't that complex anymore. It's well known, there are hundreds of thousands of people certified on it, and you can get remote install support from major HCI players for a few thousand bucks to make sure it's setup right if your really paranoid. The era of LUNs, and tuning APD/PDL timers, and figuring out SCSI queues, and Fibre Channel, and DCB is over. Even internal stuff is more safe as you have push button updates of Firmware and Controllers for the system that will roll through the cluster and take care of stuff.
Modern Hypervisor HA is actually really damn smart. IT can even detect that host MIGHT fail (failing hardware sensors) and pro-actively quarantine a host. It can fence through multiple levels (File system, host heartbeats, isolation address's) with and deliver incredibly consistent outcomes on different failure odes.
The book on HA was never that long, and Duncan's simplified it when he now gives it away for free.
Now App HA is still often expensive, and compex (relatively BAG and AD are not that bad) but for stuff like HVAC systems it's rarely an option.
-
@jaredbusch said in Looking for virtualization advice:
VMware-based appliances will probably stay with a vCenter + plug-in, but Hyper-V (and KVM soon, very soon) are getting own HTML5 GUI.
Starwind supports vVols, so you can manage storage without a plugin using that (well VASA provider is required, but no weird GUI plugin nonsense).
-
@jaredbusch said in Looking for virtualization advice:
For your setup, I would go with a single large server with local storage. Get 8 x 4 TB SATA or NL SAS disks in a RAID 10 for 16TB of usable space. The Dell R720 would work fine here (or whatever the current model is).
The R720 isn't being certified for new OS and hypervisor releases I'm pretty sure no 6.5 for it. I wouldn't get one of those.
R730 is current with R740 phasing in to replace it. I'd get the R640/740.
Don't deploy Magnetic SATA drives. I don't care if they say enterprise. Spend the extra $30 and get NL-SAS. Also worth noting is that NL-SAS have awful performance characteristics. Make sure your workload has incredibly low IO load or get ready for the database server when you run a report on the PBX to implode.
-
@jaredbusch said in Looking for virtualization advice:
@scottalanmiller said in Looking for virtualization advice:
I had to go searching for it.
Not my job. It is their job to not turn me off. Which they did. fuck that.
Everyone has public pricing if you know how to use Government contract price lists that contractually must be posted
Citrix, Microsoft, Dell, Cisco and others have public pricing, but if you just go with that you can easily miss out on a special 30% off bundle or deal (or in Cisco's case likely just overpay 70%). I get this weird attitude in IT where you don't want to deal with sales people but do you want to know the real reason why vendors don't give out price lists?
One is volume. If Bank of America agree's to buy my software I might be willing to discount to 85% for 100 million dollars in sales, because my net cost to produce licensing keys is low, and the cost to support someone that large and well developed who's going to have a homogeneous design is actually rather simple. Joe's gravel pit with 10 VM's and 2 IT guys with no formal training meanwhile could easily be a endless pit of support calls because they cook their unit that's in a trailer, and they are abusing your support organization for basic helpdesk because they don't understand basic things like NTP/DNS and routing. That's ony part of it. The other issue is how SMB's buy complicated products.
They go and buy it, and then buy the wrong damn thing. They will buy 1/2 as many nodes and not realize they need mirroring overhead (Scale shows RAW on this chart, they don't show overheads, reserves or protection. They will buy zero slack space on the boxes and crash them or not be aware of the file system formatting. They will then complain on Spiceworks "Storage vendor B sucks because their box ran out of space and caught fire and gave me measles because I didn't talk to an SE who could have helped me get the right box or realize that I shouldn't buy the product and get something else that was in budget!"As someone who worked for a VAR for years and saw customers try to design their own stuff and submit it, we ended up blocking 90% of those deals and getting a SE to get them the right stuff. Some of it was hilarious (RAID 0 ALL THE THINGS) some of it was easy to overlook stuff (Product required 3 phase power, or a rack depth they didn't have), some of it is just annoying details (Product had SFP+ ports, and they needed 10Gbase-T). Minor paperwork mistakes (Shipping address being different from install) on a customers part could cripple support responses (Dispatch part to the wrong location). Also for systems with support contracts array vendors and HCI vendors will require a diagram of how it's installed, cabled, connected so their global support teams know what you have when they call in. If you let customers size, purchase, and install none of the paperwork required for a proper support experience gets completed. A VxRAIL or Scale or Synology could be setup by a 4th grader with the check list. The reason one of them requires a trained staff member is VCE/Scale is on the hook for aggressive support and management of patching and things for the next 3-5 years, while Synology will say "meh" if you call them with an outage.
As a vendor if you shut up and take their money you risk tanking your NPS, and having them tell their friends you suck and making it harder to get deals from people who actually have budget and care about a solution actually working.
Catering to the IT know it all who thinks he can correctly purchase systems that with sub-variations often produce 1500 SKU's (RAM, CPU, Disk, NIC interfaces, Power connectors, Fan direction etc) isn't worth it to these vendors as they are more worried about other customers hearing about your negative experience than loosing a sale that was never going to happen anyways. More importantly if you correctly know the Sub-SKU's for the flex midplane option so you can run 2 HBA's in split mode on a DL380 you REALLY shouldn't be working at a customer with that skill (it's something you rarely use) and instead should be working for a VAR or distributor. You'll make a ton more money, and you'll actually get to use that skill more than once every 3-5 years.
-
@john-nicholson said in Looking for virtualization advice:
@jaredbusch said in Looking for virtualization advice:
For your setup, I would go with a single large server with local storage. Get 8 x 4 TB SATA or NL SAS disks in a RAID 10 for 16TB of usable space. The Dell R720 would work fine here (or whatever the current model is).
The R720 isn't being certified for new OS and hypervisor releases I'm pretty sure no 6.5 for it. I wouldn't get one of those.
R730 is current with R740 phasing in to replace it. I'd get the R640/740.
Don't deploy Magnetic SATA drives. I don't care if they say enterprise. Spend the extra $30 and get NL-SAS. Also worth noting is that NL-SAS have awful performance characteristics. Make sure your workload has incredibly low IO load or get ready for the database server when you run a report on the PBX to implode.
I've often seen NL-SAS coming in at $5 UNDER enterprise SATA at the same size.
-
@john-nicholson said in Looking for virtualization advice:
@jaredbusch said in Looking for virtualization advice:
@scottalanmiller said in Looking for virtualization advice:
@garyp said in Looking for virtualization advice:
We have started looking at HCI solutions, including Scale. StarWind and HPE SimpliVity as we do not the expertise in managing a hypervisor nor the time to manage it.
That's the appropriate short list. Of those, @Scale is the one that is going to offload the most from your plate. Starwind provides HC but you are still managing the hypervisor on your own, separately. It's architecturally all together, but the management console is not.
He doesn't need hyperconvergence. Don't sell him something he does not need. There is no way to intelligently get 15TB of storage on a Scale box for a reasonable price compared to local storage.
Other HCI platforms have no problem getting 15TB in a box (I had a serious discussion with an SE about a 300TB+ per host vSAN design and all the considerations around it). I think someone was taking socket based licensing to a ridiculous logical conclusion but with 100Gbps networking becoming affordable this stuff isn't that far off.
Scale doesn't have any issue with it either. I don't have large boxes at all, and mine are hybrid magnetic spinners and SSD and I still have 18.4TB per node. You can go way larger than mine. Mine are 1U, only use 75% of the drives for capacity and don't use the largest drives.
-
@john-nicholson said in Looking for virtualization advice:
@jaredbusch said in Looking for virtualization advice:
@scottalanmiller said in Looking for virtualization advice:
I had to go searching for it.
Not my job. It is their job to not turn me off. Which they did. fuck that.
Everyone has public pricing if you know how to use Government contract price lists that contractually must be posted
Someone needs to make a quick reference somewhere.
@john-nicholson said in Looking for virtualization advice:
snip very good discussion on various points of costing out solutions
VARS are important. But this project is at the very beginning of even trying to figure out which solution to narrow down to. So this is way before a VAR should be involved. This is why simple public pricing is important. The OP does not care at this point about specific deals. The OP needs to know about general costs for each solution type.
-
@jaredbusch said in Looking for virtualization advice:
@john-nicholson said in Looking for virtualization advice:
@jaredbusch said in Looking for virtualization advice:
@scottalanmiller said in Looking for virtualization advice:
I had to go searching for it.
Not my job. It is their job to not turn me off. Which they did. fuck that.
Everyone has public pricing if you know how to use Government contract price lists that contractually must be posted
Someone needs to make a quick reference somewhere.
@john-nicholson said in Looking for virtualization advice:
snip very good discussion on various points of costing out solutions
VARS are important. But this project is at the very beginning of even trying to figure out which solution to narrow down to. So this is way before a VAR should be involved. This is why simple public pricing is important. The OP does not care at this point about specific deals. The OP needs to know about general costs for each solution type.
And more importantly, about what good solutions are versus vendor trumped up ones.
-
@jaredbusch said in Looking for virtualization advice:
@john-nicholson said in Looking for virtualization advice:
@jaredbusch said in Looking for virtualization advice:
@scottalanmiller said in Looking for virtualization advice:
I had to go searching for it.
Not my job. It is their job to not turn me off. Which they did. fuck that.
Everyone has public pricing if you know how to use Government contract price lists that contractually must be posted
Someone needs to make a quick reference somewhere.
@john-nicholson said in Looking for virtualization advice:
snip very good discussion on various points of costing out solutions
VARS are important. But this project is at the very beginning of even trying to figure out which solution to narrow down to. So this is way before a VAR should be involved. This is why simple public pricing is important. The OP does not care at this point about specific deals. The OP needs to know about general costs for each solution type.
I agree, public pricing is pretty important. You can't know what VARs to engage or what approaches make sense when you don't know the relative costs of things.
