Testing oVirt...

scottalanmiller

Think of it like RAID, but spread over a network instead of locally. It's been around since 2005, and is used by some very large cloud providers today.

Not quite. Gluster is RAIN, not Network RAID. You are describing DRBD.

scottalanmiller

@stacksofplates said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee Hardware RAID is recommended because a lot of people lack the practice and skill with software RAID to appropriately maintain/troubleshoot/fix software raid.

Plus you often get features like Blind swap.

Nope...
The folks at RH say, currently, performance is better with HW RAID & that is how most of their testing/validation is done.

That's odd, but okay. I would generally think they make the recommendation because they know there is a huge lack of understanding on how software raid needs to be managed.

I'm also guessing it's because RAID isn't necessarily needed. The overhead for software RAID doesn't gain much if you're bricks are replicated anyway.

But the overhead of hardware RAID wouldn't gain anything, either.

FATeknollogee

oVirt Node is installed on a 64gb SATADOM
Gluster is on a 2TB RAID array.

Here is the disk layout:

FATeknollogee

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

Hardware RAID is recommended instead of MDADM.

By whom?

By the oVirt devs.
You can go & look for RHs best practice for RHEV, it also says use HW RAID.

stacksofplates

@scottalanmiller said in Testing oVirt...:

@stacksofplates said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee Hardware RAID is recommended because a lot of people lack the practice and skill with software RAID to appropriately maintain/troubleshoot/fix software raid.

Plus you often get features like Blind swap.

Nope...
The folks at RH say, currently, performance is better with HW RAID & that is how most of their testing/validation is done.

That's odd, but okay. I would generally think they make the recommendation because they know there is a huge lack of understanding on how software raid needs to be managed.

I'm also guessing it's because RAID isn't necessarily needed. The overhead for software RAID doesn't gain much if you're bricks are replicated anyway.

But the overhead of hardware RAID wouldn't gain anything, either.

It doesn't pull anything from the OS though. SW would actually use resources the system could use. So the only thing you would see is less capacity.

scottalanmiller

@stacksofplates said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@stacksofplates said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee Hardware RAID is recommended because a lot of people lack the practice and skill with software RAID to appropriately maintain/troubleshoot/fix software raid.

Plus you often get features like Blind swap.

Nope...
The folks at RH say, currently, performance is better with HW RAID & that is how most of their testing/validation is done.

That's odd, but okay. I would generally think they make the recommendation because they know there is a huge lack of understanding on how software raid needs to be managed.

I'm also guessing it's because RAID isn't necessarily needed. The overhead for software RAID doesn't gain much if you're bricks are replicated anyway.

But the overhead of hardware RAID wouldn't gain anything, either.

It doesn't pull anything from the OS though. SW would actually use resources the system could use. So the only thing you would see is less capacity.

Sort of, but the CPU overhead of software RAID is nominal, but as it is faster than hardware RAID and storage is the common bottleneck, it normally makes the system faster rather than slower.

So in CPU terms, there is a small overhead to software RAID that you normally can't notice. But to the overall system, there is normally a negative overhead, meaning you get more out of your server, rather than less.

scottalanmiller

@fateknollogee said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

Hardware RAID is recommended instead of MDADM.

By whom?

By the oVirt devs.
You can go & look for RHs best practice for RHEV, it also says use HW RAID.

I don't trust any recommendation that doesn't include the logic as to why. Likely their logic is "because most of our users don't know how to do MD and we don't want to be asked questions" or "we are sponsored by server vendors" or "our users always have hardware RAID already and it can't easily be disabled and they get confused" or "we think it's what people want to hear."

FreeNAS recommends software RAID, but if you dig into "why" it's based on selling more support and has nothing to do with what is good for the customer.

Kelly

@scottalanmiller said in Testing oVirt...:

@stacksofplates said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@stacksofplates said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee Hardware RAID is recommended because a lot of people lack the practice and skill with software RAID to appropriately maintain/troubleshoot/fix software raid.

Plus you often get features like Blind swap.

Nope...
The folks at RH say, currently, performance is better with HW RAID & that is how most of their testing/validation is done.

That's odd, but okay. I would generally think they make the recommendation because they know there is a huge lack of understanding on how software raid needs to be managed.

I'm also guessing it's because RAID isn't necessarily needed. The overhead for software RAID doesn't gain much if you're bricks are replicated anyway.

But the overhead of hardware RAID wouldn't gain anything, either.

It doesn't pull anything from the OS though. SW would actually use resources the system could use. So the only thing you would see is less capacity.

Sort of, but the CPU overhead of software RAID is nominal, but as it is faster than hardware RAID and storage is the common bottleneck, it normally makes the system faster rather than slower.

So in CPU terms, there is a small overhead to software RAID that you normally can't notice. But to the overall system, there is normally a negative overhead, meaning you get more out of your server, rather than less.

I am curious (not arguing, but wanting to learn) what resources you have that show a performance comparison between software and hardware RAID.

scottalanmiller

@kelly said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@stacksofplates said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@stacksofplates said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

@dustinb3403 said in Testing oVirt...:

@fateknollogee Hardware RAID is recommended because a lot of people lack the practice and skill with software RAID to appropriately maintain/troubleshoot/fix software raid.

Plus you often get features like Blind swap.

Nope...
The folks at RH say, currently, performance is better with HW RAID & that is how most of their testing/validation is done.

That's odd, but okay. I would generally think they make the recommendation because they know there is a huge lack of understanding on how software raid needs to be managed.

I'm also guessing it's because RAID isn't necessarily needed. The overhead for software RAID doesn't gain much if you're bricks are replicated anyway.

But the overhead of hardware RAID wouldn't gain anything, either.

It doesn't pull anything from the OS though. SW would actually use resources the system could use. So the only thing you would see is less capacity.

Sort of, but the CPU overhead of software RAID is nominal, but as it is faster than hardware RAID and storage is the common bottleneck, it normally makes the system faster rather than slower.

So in CPU terms, there is a small overhead to software RAID that you normally can't notice. But to the overall system, there is normally a negative overhead, meaning you get more out of your server, rather than less.

I am curious (not arguing, but wanting to learn) what resources you have that show a performance comparison between software and hardware RAID.

I'll have to research, but I know that it was standard certification material that 2000 was the inflection year when normal servers software RAID pulled ahead of hardware RAID. It was the Pentium IIIs processor that made the difference. It was the first time that the average server had so much space CPU capacity that using a trivial amount for software RAID didn't affect the main system, and that the main CPUs were so dramatically fast compared to the cheap chips doing the hardware offloading that it made software generally faster than hardware.

This is also why hardware RAID is exclusive to the IA32 world. The big RISC server processors like Power, Sparc, PA-RISC and Alpha from the 1990s never had, and still don't, have hardware RAID options because they had excess CPU power and software RAID was faster on them from day one. So hardware RAID just never became a thing in those markets. It started exclusively because the old Intel chips prior to the P3s (that is not plural, it is the "S" model) were so underpowered.

FATeknollogee

Either way, at the end of the day, RAID controllers are like cars...
People go with what they're comfortable with....you can argue (aka debate) the merits of one vs. the other.
Me, I could go either way.

scottalanmiller

@fateknollogee said in Testing oVirt...:

Either way, at the end of the day, RAID controllers are like cars...
People go with what they're comfortable with....you can argue (aka debate) the merits of one vs. the other.
Me, I could go either way.

They shouldn't be treated that way. Each has a place and each should be used where appropriate. Not much in IT should come down to opinion, it should be "appropriate addressing of the business case."

FATeknollogee

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

Either way, at the end of the day, RAID controllers are like cars...
People go with what they're comfortable with....you can argue (aka debate) the merits of one vs. the other.
Me, I could go either way.

They shouldn't be treated that way. Each has a place and each should be used where appropriate. Not much in IT should come down to opinion, it should be "appropriate addressing of the business case."

You are right, it shouldn't but I live in the real world & that's just how it goes.
You aren't going to save the planet :grinning_face_with_smiling_eyes:

FATeknollogee

Gluster Storage Network:
bond 0 is 2 x 10gbE interfaces - Mode 6 Adaptive Load Balancing (balance-alb)

eno1:
ovirtmgmt - Management network

eno2:
Virtual machine traffic

scottalanmiller

@fateknollogee said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

Either way, at the end of the day, RAID controllers are like cars...
People go with what they're comfortable with....you can argue (aka debate) the merits of one vs. the other.
Me, I could go either way.

They shouldn't be treated that way. Each has a place and each should be used where appropriate. Not much in IT should come down to opinion, it should be "appropriate addressing of the business case."

You are right, it shouldn't but I live in the real world & that's just how it goes.
You aren't going to save the planet :grinning_face_with_smiling_eyes:

No, but I can save a lot of IT pros.

coliver

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

Either way, at the end of the day, RAID controllers are like cars...
People go with what they're comfortable with....you can argue (aka debate) the merits of one vs. the other.
Me, I could go either way.

They shouldn't be treated that way. Each has a place and each should be used where appropriate. Not much in IT should come down to opinion, it should be "appropriate addressing of the business case."

You are right, it shouldn't but I live in the real world & that's just how it goes.
You aren't going to save the planet :grinning_face_with_smiling_eyes:

No, but I can save a lot of businesses from "IT pros".

FTFY.

JaredBusch

@coliver said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

@scottalanmiller said in Testing oVirt...:

@fateknollogee said in Testing oVirt...:

Either way, at the end of the day, RAID controllers are like cars...
People go with what they're comfortable with....you can argue (aka debate) the merits of one vs. the other.
Me, I could go either way.

They shouldn't be treated that way. Each has a place and each should be used where appropriate. Not much in IT should come down to opinion, it should be "appropriate addressing of the business case."

You are right, it shouldn't but I live in the real world & that's just how it goes.
You aren't going to save the planet :grinning_face_with_smiling_eyes:

No, but I can save a lot of businesses from "IT pros".

FTFY.

Hahahahah

FATeknollogee

Starting to test some backups of vm's on oVirt & standalone KVM:

vProtect 3.6

oVirt VMs:

Standalone KVM VMs:

Romo

@fateknollogee Free for standalone KVM hosts, this seems interesting. Please share your testing results =).

scottalanmiller

@romo we've got several places where we might want to use that, and more soon.

JaredBusch

@scottalanmiller said in Testing oVirt...:

@romo we've got several places where we might want to use that, and more soon.

Why? Agentless VM Level backups are useless. You've said that often enough.