ServerBear Specs on Scale HC3
-
@aaronstuder said:
Couldn't I just create a XenServer Cluster with Xen Orchestra, and get the same thing?
Nowhere close, I'm afraid. The thing that makes the Scale cluster important is the RAIN based scale out RLS system on which it is built. XS with XO would give you the same basic "single pane of glass" interface stuff, but you aren't getting a Scale for the interface. It's the storage mostly and the HA management secondary that make it valuable.
The RAIN storage here mirrors at the block level across the cluster providing a very high durability storage layer. And very importantly that's a native storage layer, in the kernel. There is no VSA here, this is a more advanced and more powerful approach. The storage layer runs right in the hypervisor kernel.
Then on top of that, there is integrated storage and compute management, so both layers know what the other layer is doing. Performance, faults, capacity data is all transparent between the two. So this provides for a level of storage performance, scale out and reliability that you cannot easily replicate on your own.
-
Now assuming that you do want to take the "poor man's" approach and try to build something like this on your own, of course there are tools for that. Using either KVM or Xen you can add a scale out storage layer to that. The two key ones on the market are Gluster and CEPH. You would have to build that component yourself. DRBD is great for two nodes but is not scale out, it's a good product but a different animal here. So you need to build your own durable, HA scale out storage layer on which to run Xen or KVM. Then layer the HA and management on top of that.
-
@scottalanmiller said:
hich it is built. XS with XO would give you the same basic "single pane of glass" interface stuff, but you aren't getting a Scale for
How do you have three nodes and only loose 50% storage, yet loose nothing when a node fails?
-
@Dashrender said:
@scottalanmiller said:
hich it is built. XS with XO would give you the same basic "single pane of glass" interface stuff, but you aren't getting a Scale for
How do you have three nodes and only loose 50% storage, yet loose nothing when a node fails?
RAIN mirroring
In RAID terms, think of network RAID 1+. -
The blocks are mirrored. No matter what you write to one node, it is replicated to at least one additional node. But no node is a "pair".
-
@scottalanmiller said:
@Dashrender said:
@scottalanmiller said:
hich it is built. XS with XO would give you the same basic "single pane of glass" interface stuff, but you aren't getting a Scale for
How do you have three nodes and only loose 50% storage, yet loose nothing when a node fails?
RAIN mirroring
In RAID terms, think of network RAID 1+.Is RAIN mirroring always 50% available? I know I know.. mirroring kinda implies that, but I have to ask the question anyway.
Also, what manages this? I assume this management is also mirrored out over the nodes?
-
@Dashrender said:
Is RAIN mirroring always 50% available? I know I know.. mirroring kinda implies that, but I have to ask the question anyway.
Yes, mirroring is always 50%. RAIN is not always mirroring.
-
@Dashrender said:
Also, what manages this? I assume this management is also mirrored out over the nodes?
Each node has a completely independent management system. You can log into any of the nodes to manage the cluster. It's a multi-master system. So fully HA there, not even a blip if a node goes down (unless you are looking at the interface for that specific node.)
-
@scottalanmiller said:
The RAIN storage here mirrors at the block level across the cluster providing a very high durability storage layer. And very importantly that's a native storage layer, in the kernel. There is no VSA here, this is a more advanced and more powerful approach. The storage layer runs right in the hypervisor kernel.
A couple of my colleagues told me they lost TWO nodes in their 4 node scale cluster... and everything kept right on trucking... and these units are the 3-year old models.
-
@dafyre said:
@scottalanmiller said:
The RAIN storage here mirrors at the block level across the cluster providing a very high durability storage layer. And very importantly that's a native storage layer, in the kernel. There is no VSA here, this is a more advanced and more powerful approach. The storage layer runs right in the hypervisor kernel.
A couple of my colleagues told me they lost TWO nodes in their 4 node scale cluster... and everything kept right on trucking... and these units are the 3-year old models.
That can easily work as long as you don't use your capacity all the way up. If you lose them one at a time and it has time to rebalance you would be all set in any way. If you lose two at exactly the same time, it's a bit more of luck
-
@mlnews said:
@dafyre said:
@scottalanmiller said:
The RAIN storage here mirrors at the block level across the cluster providing a very high durability storage layer. And very importantly that's a native storage layer, in the kernel. There is no VSA here, this is a more advanced and more powerful approach. The storage layer runs right in the hypervisor kernel.
A couple of my colleagues told me they lost TWO nodes in their 4 node scale cluster... and everything kept right on trucking... and these units are the 3-year old models.
That can easily work as long as you don't use your capacity all the way up. If you lose them one at a time and it has time to rebalance you would be all set in any way. If you lose two at exactly the same time, it's a bit more of luck
It was a whole lotta luck for them, lol. They're not running at full capacity, so they were safe.
-
@dafyre said:
@mlnews said:
@dafyre said:
@scottalanmiller said:
The RAIN storage here mirrors at the block level across the cluster providing a very high durability storage layer. And very importantly that's a native storage layer, in the kernel. There is no VSA here, this is a more advanced and more powerful approach. The storage layer runs right in the hypervisor kernel.
A couple of my colleagues told me they lost TWO nodes in their 4 node scale cluster... and everything kept right on trucking... and these units are the 3-year old models.
That can easily work as long as you don't use your capacity all the way up. If you lose them one at a time and it has time to rebalance you would be all set in any way. If you lose two at exactly the same time, it's a bit more of luck
It was a whole lotta luck for them, lol. They're not running at full capacity, so they were safe.
One of the nice things about scale out is that it also often means "scale back". So if you are not over using the system, it can rebalance as you fail back to a smaller system.
You can also replicate entire clusters for even more reliability.
-
The more I read about Scale and peoples views the more I want it!!!!!!
Now to just get the board and management to think we need to refresh our hardware lol
-
I guess I just don't get it. This is basically just Dell Servers running via a VSAN connected over 10GB networking. I guess I don't see what's so special about it.
-
@aaronstuder said:
I guess I just don't get it. This is basically just Dell Servers running via a VSAN connected over 10GB networking. I guess I don't see what's so special about it.
It's not the hardware that makes it special. It's the software they run behind it all. The storage provides similar features to VSAN, except you have the ability to simply add another node to expand storage. I'm not sure how that works with VSAN... but with Scale, all you have to buy is another node... no extra software licenses or anything like that.
-
@aaronstuder said:
I guess I just don't get it. This is basically just Dell Servers running via a VSAN connected over 10GB networking. I guess I don't see what's so special about it.
Well that alone would be pretty special as it is far cheaper than running VSAN. VSAN and the full VMware stack is a direct competitor. Scale is older than VSAN. VSAN is SAN based, though, Scale is not. This is fully direct IO, no extra SAN layer. The kernel talks directly to the storage, no translation.
-
@dafyre said:
It's not the hardware that makes it special. It's the software they run behind it all. The storage provides similar features to VSAN, except you have the ability to simply add another node to expand storage. I'm not sure how that works with VSAN... but with Scale, all you have to buy is another node... no extra software licenses or anything like that.
And it "just expands". Really easily to grow. And tiering is right around the corner.
-
Scott, you mentioned that this is all at the kernel level - could you roll your own version of this? Of course you wouldn't have their pretty interface, but could you build this yourself, like a person building XO themself?
-
@Dashrender said:
Scott, you mentioned that this is all at the kernel level - could you roll your own version of this?
Sure, you'd have to write your own storage layer, though. So it's not trivial in any way.
-
@scottalanmiller said:
@Dashrender said:
Scott, you mentioned that this is all at the kernel level - could you roll your own version of this?
Sure, you'd have to write your own storage layer, though. So it's not trivial in any way.
OH.. that's where I was confused I guess... I thought the storage layer was part of KVM (that's the hypervisor they use, right?)
