Follow up on Hyper-V High availability? or only VMware
-
Certainly for most use cases Hyper-V Replica is enough. The question here would be, is the effort of Starwind enough to make the benefits worth it. My vote is for yes, it protects not just against downtime and small data loss but also against data corruption that, while unlikely, can happen.
-
@scottalanmiller said:
With that range of IOPS, it could be easily worth it to invest in another four drives, two going to each machine, to move that RAID 10 array from four to six spindles. That's a 50% increase in performance over what you have now for probably relatively little money. The CPU and memory are overkill for your needs, but the disks are a bottleneck.
Can you go over the bottleneck part? How can I tackle this to improve performance? We are 85/15 R/W
-
@JaredBusch said:
@scottalanmiller said:
I think that Starwind RLS and Hyper-V and RAID 10 is all perfect for this. Fast, good use of what you have, extremely safe.
Why even add in starwind? There is no use case for here for anything beyond local storage and built in replication for failover.
Because HA is mentioned in the title, even though it's not mentioned in the OP. Also it was mentioned in the old thread.
If not for those things, I too would say that StarWind would be overkill.
-
My question is - Are you still running VMWare, or have you moved to Hyper-V?
Why are you running two servers instead of just one? The VM's are completely separate, so splitting them over two pieces of hardware really doesn't gain you anything, other than higher electricity bills, higher cooling costs, more HDs to purchase. More warranty to purchase. etc etc
-
@Dashrender said:
My question is - Are you still running VMWare, or have you moved to Hyper-V?
Why are you running two servers instead of just one? The VM's are completely separate, so splitting them over two pieces of hardware really doesn't gain you anything, other than higher electricity bills, higher cooling costs, more HDs to purchase. More warranty to purchase. etc etc
The 2 servers came into play mainly for vSAN. If my understanding is correct: vSAN required identical storage capacity on at least 2 nodes. I even want to create a cluster where 2 VMs per host (4 VMs total). but my boss... told me he wants them separate. He's old school and never heard of VM until 2 months ago when I brought it up. To him VM is very new.. even it has bee naround for years.
-
It's to bad you can't show him and have him understand/believe that this stuff is a decade+ old (for VMWare) and 8 or so years old for Hyper-V. Definitely not new.
Moving onto Scott's licensing comment.
Assuming you have two server licenses today, unless you want to purchase additional licenses, you should only create one VM on each host. This will allow you to remain fully licensed, legally. Why? Because, since each license allows you to have a 2 VMs, you'll never have more than two VMs per host, so you'll be covered.
-
@Dashrender said:
It's to bad you can't show him and have him understand/believe that this stuff is a decade+ old (for VMWare) and 8 or so years old for Hyper-V. Definitely not new.
Moving onto Scott's licensing comment.
Assuming you have two server licenses today, unless you want to purchase additional licenses, you should only create one VM on each host. This will allow you to remain fully licensed, legally. Why? Because, since each license allows you to have a 2 VMs, you'll never have more than two VMs per host, so you'll be covered.
Licensing is quite confusing per wording. But, according to Microsoft License agreement, 1 license cover [ 1 host + 2 vm on the same host] it does not cover 2 vms on separate host.
Source: Virtual Machine Guest Licensing and Hyper-V (2012 & 2012 R2)
I did not read the whole article just skim through it. -
@LAH3385 said:
@Dashrender said:
It's to bad you can't show him and have him understand/believe that this stuff is a decade+ old (for VMWare) and 8 or so years old for Hyper-V. Definitely not new.
Moving onto Scott's licensing comment.
Assuming you have two server licenses today, unless you want to purchase additional licenses, you should only create one VM on each host. This will allow you to remain fully licensed, legally. Why? Because, since each license allows you to have a 2 VMs, you'll never have more than two VMs per host, so you'll be covered.
Licensing is quite confusing per wording. But, according to Microsoft License agreement, 1 license cover [ 1 host + 2 vm on the same host] it does not cover 2 vms on separate host.
Source: Virtual Machine Guest Licensing and Hyper-V (2012 & 2012 R2)
I did not read the whole article just skim through it.This is correct. As long as you only use that host as a hypervisor and Dom0 and have no other roles installed on it.
-
That's correct - and I might even be wrong in my belief that you'd be good to go in my presented solution.
Scott, what do you think? Would he need a total of 4 licenses no matter what? The two for the normal location of the VM, and two for the failover location?
Or because they are VM's, would he be able to failover the VM to the open VM license spot on the other server.
I tend to think my solution works because of how Datacenter licensing works. DC allows you unlimited VMs, so you have unlimited slots open to accept an existing VM from another host into this one.
Let's look at this another way.
Let's say I had 5 VM hosts, each with one VM (for example purposes only) I have a 6th VM host that is a BIG Boy. I setup the first 5 VM hosts to fail over only to the 6th. I believe that I could purchase DC license only for the 6th host, and standard license for the others and be fine.
-
@Dashrender said:
That's correct - and I might even be wrong in my belief that you'd be good to go in my presented solution.
Scott, what do you think? Would he need a total of 4 licenses no matter what? The two for the normal location of the VM, and two for the failover location?
Or because they are VM's, would he be able to failover the VM to the open VM license spot on the other server.
I tend to think my solution works because of how Datacenter licensing works. DC allows you unlimited VMs, so you have unlimited slots open to accept an existing VM from another host into this one.
Let's look at this another way.
Let's say I had 5 VM hosts, each with one VM (for example purposes only) I have a 6th VM host that is a BIG Boy. I setup the first 5 VM hosts to fail over only to the 6th. I believe that I could purchase DC license only for the 6th host, and standard license for the others and be fine.
I'll wait for Scott to give a better answer but here is my point of view:
Standard and DC is pretty much the same license as far as Host goes. Host can only have 1 role and that is Hyper-V. However, Standard license allow to be use on 2 VMs that is on the same Host. DC, on the other hand, can be use unlimited time as resources allowed.. but it has to be on the same host as DC.
So the real question is how many VMs will be on the BIG Boy? 2 or less = Standard. 2+ then DC.
That's my understanding. Would love if someone to clarify this as well.There really need to be an infograph on Microsoft licensing. I have to read EULA couple of itmes to grasp the concept
-
@LAH3385 said:
@scottalanmiller said:
With that range of IOPS, it could be easily worth it to invest in another four drives, two going to each machine, to move that RAID 10 array from four to six spindles. That's a 50% increase in performance over what you have now for probably relatively little money. The CPU and memory are overkill for your needs, but the disks are a bottleneck.
Can you go over the bottleneck part? How can I tackle this to improve performance? We are 85/15 R/W
Bottleneck: Your server is doing around 200 IOPS, my desktop does around 100,000 IOPS
In nearly any server, it is the storage that is the part that will slow you down. Disks are super slow compared to CPU, memory and other components. So this is where you tend to invest to really speed things up.
RAID 10 arrays increase in performance linearly (for all intents and purposes.) So by moving from four disks to six disks in the array we got 50% IOPS. Going from four to eight would give us a 100% boost! Moving to faster drives gives us a big boost too. Going from 7200 RPM to 10K RPM is a nearly 50% increase again. Going from SATA to SAS makes drives more efficient, often to the 5 - 15% range.
RAID controllers have a cache (normally) that can do a lot to improve performance, especially of writes. And SSDs and SSD Caches can take us into completely different performance categories.
-
@Dashrender said:
If not for those things, I too would say that StarWind would be overkill.
You did not feel that way in the thread talking about Veeam async replication versus Starwind on VMware in the other thread yesterday. Why do you feel one way there and another here, only because one is a single vendor and the other is always multiple? What is the factor causing the change of opinion since the design and architectures remains constant.
-
@LAH3385 said:
To him VM is very new.. even it has bee naround for years.
How out of touch is he? VMs have been around since 1964 and the industry standard for critical workloads since the 1990s and the best practice even in the SMB since it was feasible around 2005.
-
@scottalanmiller said:
@LAH3385 said:
@scottalanmiller said:
With that range of IOPS, it could be easily worth it to invest in another four drives, two going to each machine, to move that RAID 10 array from four to six spindles. That's a 50% increase in performance over what you have now for probably relatively little money. The CPU and memory are overkill for your needs, but the disks are a bottleneck.
Can you go over the bottleneck part? How can I tackle this to improve performance? We are 85/15 R/W
Bottleneck: Your server is doing around 200 IOPS, my desktop does around 100,000 IOPS
In nearly any server, it is the storage that is the part that will slow you down. Disks are super slow compared to CPU, memory and other components. So this is where you tend to invest to really speed things up.
RAID 10 arrays increase in performance linearly (for all intents and purposes.) So by moving from four disks to six disks in the array we got 50% IOPS. Going from four to eight would give us a 100% boost! Moving to faster drives gives us a big boost too. Going from 7200 RPM to 10K RPM is a nearly 50% increase again. Going from SATA to SAS makes drives more efficient, often to the 5 - 15% range.
RAID controllers have a cache (normally) that can do a lot to improve performance, especially of writes. And SSDs and SSD Caches can take us into completely different performance categories.
I am planning on getting some SSD next year Q2/Q3 for some application or vSAN cache.
I asked the question here Still waiting for xByte to reply -
WIth virtualization you license Microsoft Server in terms of capacity of that host for windows VMs. If you have a typical host(2 or less processors) and the most VMs you will ever run on it in any situation(production, failover, maintenance, etc) is 2vm's, you will need 1 Standard Server License for that host. You have to apply that to all hosts. You are no longer licensing VMs, you are licensing what the host can run on it.
-
@scottalanmiller said:
@LAH3385 said:
To him VM is very new.. even it has bee naround for years.
How out of touch is he? VMs have been around since 1964 and the industry standard for critical workloads since the 1990s and the best practice even in the SMB since it was feasible around 2005.
He graduate with Computer Science and started his own company ever since. I could safely say he has been in a personal management position for the past 25 years or so.. with little to no involvement in IT management.
He's the CEO if that make any different -
@LAH3385 said:
Licensing is quite confusing per wording. But, according to Microsoft License agreement, 1 license cover [ 1 host + 2 vm on the same host] it does not cover 2 vms on separate host.
The host bit is technically correct but effectively confusing. HyperV is free and should not be installed in conjunction with Windows Server. So while there is a license that lets you do this, it's not a best practice or necessary so adds a whole world of complication that is best avoided.
-
Moving to DC licensing is extremely expensive. You don't move there just in a 2+ situation, you move there in a 8 or maybe 10 or more VM situation.
You can have a single VM host and have 6 VMs on it, requiring you to purchase 3 standard licenses for that host. That would cost you roughly $2400 ($800 a license).
DC = $5600 or the same as 7 standard licenses, which would give you 14 VMs.
But, in a failover situation if you have 7 VMs on one host, and 7 VMs on another host and they failover to each other, assuming a full failure situation, you'd have 14 VMs on a single host.. and therefore your best cost situation is a DC license on each.
But if you only have 6 VMs on each server, it will cost you less to purchase 24 standard server licenses (12 * $800 = $9600) instead of buying 2 DC licenses for $11,200 (2 x $5600).
But you have an unusual situation, though it would really be the same for any case where you have fewer VMs that DC makes sense, AND you have an odd number of VMs on each host.
having the odd number of VMs presents the question I asked above.
-
@LAH3385 said:
He graduate with Computer Science and started his own company ever since.
So if he has no experience or training in IT, why is he inserting an opinion? Comp Sci would teach him nothing at all about IT. Although even first year comp sci students at community colleges were required to do virtualization so he's very, very out of touch even for entry level comp sci work.
[Source: NY's Monroe Community College required writing a functional virtualization platform in its very first entry level programming class as the first project. That's freshman level work at a two year community college. This is circa 1997.]
-
@scottalanmiller said:
@Dashrender said:
If not for those things, I too would say that StarWind would be overkill.
You did not feel that way in the thread talking about Veeam async replication versus Starwind on VMware in the other thread yesterday. Why do you feel one way there and another here, only because one is a single vendor and the other is always multiple? What is the factor causing the change of opinion since the design and architectures remains constant.
I would be assuming an architectural change, mainly that you would have fewer disks in each host to save on that expense, on the assumption that you had to maintain two servers.
But if all other hardware is staying the same, then sure, absolutely toss StarWind in there. You take a hit on performance, but it's probably one you can afford.
