Food for thought: Fixing an over-engineered environment
-
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
@travisdh1 said in Food for thought: Fixing an over-engineered environment:
@eddiejennings Getting there. Hyper-V should be installed on all the servers. Even when you just have a single VM running on a hypervisor, the time savings in restoring after a hardware failure is worthwhile.
Server 3 could be you're testing/development area and with Hyper-V installed on all of them, you could quickly restore images to it if one of the other servers goes down. Yeah, the performance would just hurt, but at least you'd be running till actual hardware replacements arrive.
I see. So I'd also make Server 1 a hypervisor with one VM, and that VM would provide block storage to the VM that would be running backup software. That make sense; since on second thought, there would be no reason not to just make it a hypervisor.
Yep.
Where you'd see a major benefit in this case is if you ever needed to restore the backup storage from another backup. The hardware doesn't matter so long as it has the needed amount of storage.
-
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
@travisdh1 said in Food for thought: Fixing an over-engineered environment:
@eddiejennings Getting there. Hyper-V should be installed on all the servers. Even when you just have a single VM running on a hypervisor, the time savings in restoring after a hardware failure is worthwhile.
Server 3 could be you're testing/development area and with Hyper-V installed on all of them, you could quickly restore images to it if one of the other servers goes down. Yeah, the performance would just hurt, but at least you'd be running till actual hardware replacements arrive.
I see. So I'd also make Server 1 a hypervisor with one VM, and that VM would provide block storage to the VM that would be running backup software. That make sense; since on second thought, there would be no reason not to just make it a hypervisor.
Yup, basically no exception to installing the hypervisor first.
-
@travisdh1 said in Food for thought: Fixing an over-engineered environment:
@eddiejennings Getting there. Hyper-V should be installed on all the servers. Even when you just have a single VM running on a hypervisor, the time savings in restoring after a hardware failure is worthwhile.
Can add stability, too.
-
This is a disaster and needs updated from what you learned today in your other thread.
But beyond that, you talk about putting the connections to the router. This is wrong. A router routes traffic. it is not a switch.
You still require a switch.
You put multiple NICS in a team and plug those into the switch. If the switch supports full LACP then you can get awesome performance, if not, switch independent mode is the best solution.
You install Hyper-V Server 2016 on all three boxes.
On all Servers:
Create 1 partition for the Hyper-V Server drive C (I use 80GB, but I think the technical minimum is 32GB)
Create 1 partition from the rest of the space to mount as drive D inside Hyper-V
This D drive is where all of the guest files will be stored. Config files as well as replicas, checkpoints (snapshots), and the virtual hard disks.On Server 2, restore all your current servers as new VMs
On Server 1, create a small VHDX to install Windows and run Veeam.
On Server 1 create a large VHDX to house the backups. This will bethe D drive inside the Veeam guest.On Server 3, setup a test environment or sell the hardware. You could use Hyper-V replication, but you need SA on the original guest VMs or full licenses for the replicas.
-
@jaredbusch Not as disastrous as what we have, but not good either. This thread is my brainstorm thread :D. I'll put an updated diagram up tomorrow, when I return to work.
On the switch, you're right. Even though the router can probably handle the traffic, it is better to let it do its job and the switch likewise. Also since I have plenty of NICs on each physical server, I can utilizing teaming as you suggested -- which is what we have now, but with the needless VLANs. The Dell switch I have supports LACP, which is what we're using for the current teams.
On the storage configuration of the servers (drive Cs and Ds), that's what I was considering. I'm glad you mentioned putting the config files, etc., on the same partition as the VHDs, as I think about it, I don't see any advantage to keeping the VHDs separate from everything else.
From my OP, our main dev / one of my bosses (yes, that's as screwy as it sounds) is envisioning this idea of eventually having staging VMs, which could perhaps create a use for server 3. My main focus now is fixing the terrible environment from the OP.
-
Diagram update following lessons in backup design.
One other thing for me to consider is that each of these physical servers has 8 NICs (excluding the IPMI NIC). That's of course way overkill, but since I have them , would there be a reason to not team more than 4 NICs together (notwithstanding the fact that a decision hasn't been made yet about what to do with Server 3)?
-
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
That's of course way overkill, but since I have them , would there be a reason to not team more than 4 NICs together (notwithstanding the fact that a decision hasn't been made yet about what to do with Server 3)?
Four is the max you can consider in a load balancing team. If you move to pure failover, you can do unlimited. Beyond four, the algorithms become so inefficient that you don't get faster, and by six, you start actually getting slower. Most people only go to two, four is the absolute max to consider. Since you have eight (how did that happen?) you might as well do four. But the rest are wasted or could be used for a different network connection entirely.
-
@scottalanmiller said in Food for thought: Fixing an over-engineered environment:
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
That's of course way overkill, but since I have them , would there be a reason to not team more than 4 NICs together (notwithstanding the fact that a decision hasn't been made yet about what to do with Server 3)?
Four is the max you can consider in a load balancing team. If you move to pure failover, you can do unlimited. Beyond four, the algorithms become so inefficient that you don't get faster, and by six, you start actually getting slower. Most people only go to two, four is the absolute max to consider. Since you have eight (how did that happen?) you might as well do four. But the rest are wasted or could be used for a different network connection entirely.
Wouldn't this be 4 max per vNetwork in the VM host?
-
If you need more bandwidth than 4 GB, it might be time to look at 10 GB connections.
-
@dashrender said in Food for thought: Fixing an over-engineered environment:
If you need more bandwidth than 4 GB, it might be time to look at 10 GB connections.
Where "might be" = "long past due."
-
@dashrender said in Food for thought: Fixing an over-engineered environment:
@scottalanmiller said in Food for thought: Fixing an over-engineered environment:
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
That's of course way overkill, but since I have them , would there be a reason to not team more than 4 NICs together (notwithstanding the fact that a decision hasn't been made yet about what to do with Server 3)?
Four is the max you can consider in a load balancing team. If you move to pure failover, you can do unlimited. Beyond four, the algorithms become so inefficient that you don't get faster, and by six, you start actually getting slower. Most people only go to two, four is the absolute max to consider. Since you have eight (how did that happen?) you might as well do four. But the rest are wasted or could be used for a different network connection entirely.
Wouldn't this be 4 max per vNetwork in the VM host?
Correct, if the connects are independent, you get to do another four.
-
@dashrender said in Food for thought: Fixing an over-engineered environment:
If you need more bandwidth than 4 GB, it might be time to look at 10 GB connections.
I don't need more than 1 GB judging from what New Relic has shown me; however, since I have the hardware (and 4 of the 8 NICs are integrated on the motherboard) I might as well configure it to give the most performance it can.
-
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
@dashrender said in Food for thought: Fixing an over-engineered environment:
If you need more bandwidth than 4 GB, it might be time to look at 10 GB connections.
I don't need more than 1 GB judging from what New Relic has shown me; however, since I have the hardware (and 4 of the 8 NICs are integrated on the motherboard) I might as well configure it to give the most performance it can.
That's not how things work. Teaming is for bandwidth, not-teaming is for latency. Working in banks, we specifically avoided teaming because it increases latency slowing down the network traffic on a per packet basis. Everything is a trade off, or there wouldn't be options.
It's like adding more memory to your server. It's more stuff that can go in memory, but more memory that the CPU has to manage and therefore, it adds load to the server which turns into latency for processes.
-
@scottalanmiller said in Food for thought: Fixing an over-engineered environment:
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
@dashrender said in Food for thought: Fixing an over-engineered environment:
If you need more bandwidth than 4 GB, it might be time to look at 10 GB connections.
I don't need more than 1 GB judging from what New Relic has shown me; however, since I have the hardware (and 4 of the 8 NICs are integrated on the motherboard) I might as well configure it to give the most performance it can.
That's not how things work. Teaming is for bandwidth, not-teaming is for latency. Working in banks, we specifically avoided teaming because it increases latency slowing down the network traffic on a per packet basis. Everything is a trade off, or there wouldn't be options.
It's like adding more memory to your server. It's more stuff that can go in memory, but more memory that the CPU has to manage and therefore, it adds load to the server which turns into latency for processes.
That makes sense. Performance was a poor choice of words.
-
I never use IPMI.
-
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
@dashrender said in Food for thought: Fixing an over-engineered environment:
If you need more bandwidth than 4 GB, it might be time to look at 10 GB connections.
I don't need more than 1 GB judging from what New Relic has shown me; however, since I have the hardware (and 4 of the 8 NICs are integrated on the motherboard) I might as well configure it to give the most performance it can.
This is not only bad for the reasons Scott said, but it's also a waste of Switch ports and resources.
If you only need 1 Gb, then I'd remove the card (less power use) and only use two onboard NICs.
-
This is not only bad for the reasons Scott said, but it's also a waste of Switch ports and resources.
If you only need 1 Gb, then I'd remove the card (less power use) and only use two onboard NICs.
The whole situation is a waste of resources. I'm looking to see how to best utilize them.
-
@jaredbusch said in Food for thought: Fixing an over-engineered environment:
I never use IPMI.
@JaredBusch thought IPMI was something special for Hyper-V, not that you were talking about the iDRAC like interface - he stands corrected and uses the iDRAC like interface as much as he can.
-
@eddiejennings said in Food for thought: Fixing an over-engineered environment:
This is not only bad for the reasons Scott said, but it's also a waste of Switch ports and resources.
If you only need 1 Gb, then I'd remove the card (less power use) and only use two onboard NICs.
The whole situation is a waste of resources. I'm looking to see how to best utilize them.
Right, so for this part, the best would likely be two 1 Gb (on board) NICs in a team.
-
@jaredbusch said in Food for thought: Fixing an over-engineered environment:
I never use IPMI.
I've been underwhelmed with it. If you're curious, this is the motherboard that's on all of these servers:
