XenServer, local storage, and redundancy/backups
-
@Reid-Cooper said:
@Kelly You would have to have a SAS controller of some type in there for the drives that are attached for CEPH now.
There is an onboard controller, but it isn't running any RAID configuration that I can tell.
-
@Kelly said:
There is an onboard controller, but it isn't running any RAID configuration that I can tell.
It would not be for CEPH. CEPH is a RAIN system, there would be no RAID. But what it was doing isn't an issue. What we care about going forward is what we can do. The SAS controller has the drives attached and that's all that we would care about when looking at the software RAID from MD. The SAS controller isn't what provides the RAID, it is just what attaches the drives.
-
So, to sum up, the recommendation would be to run XS independently on each of the four, configuring RAID 1 or 10, if possible, and then use XO to manage it all? Is that correct?
-
@Kelly said:
So, to sum up, the recommendation would be to run XS independently on each of the four, configuring RAID 1 or 10, if possible, and then use XO to manage it all? Is that correct?
That is what I am thinking. Or RAID 6 for more capacity. With eight drives, RAID 6 might be a decent option.
-
How much storage do you have in CEPH now? what kind of resiliency do you have today? I'm completely unfamiliar with CEPH, if you loose a drive, I'm assuming you don't loose an entire array?
-
@Dashrender said:
How much storage do you have in CEPH now? what kind of resiliency do you have today? I'm completely unfamiliar with CEPH, if you loose a drive, I'm assuming you don't loose an entire array?
CEPH is RAIN. Very advanced. Compare to Gluster, Luster, Exablox or Scale's storage.
-
@Dashrender said:
How much storage do you have in CEPH now? what kind of resiliency do you have today? I'm completely unfamiliar with CEPH, if you loose a drive, I'm assuming you don't loose an entire array?
It is a very cool tool. As @scottalanmiller said, it is a RAIN (hadn't heard that term before yesterday). Basically it is software that will write to multiple nodes (it is even slow link aware) and enable you to convert commodity hardware into a resilient storage system. I would consider keeping it if it were not able to coexist (as near as I can tell) with XS.
As for total storage it is pretty low. Each host is running at < 20 TB in absolute terms. Since Ceph requires three writes that means I'm getting quite a bit less than this on average.
-
I am pretty sure that CEPH and Xen can coexist, but I don't know about with XenServer. Doing so would likely be very awkward at best.
-
@StrongBad said:
I am pretty sure that CEPH and Xen can coexist, but I don't know about with XenServer. Doing so would likely be very awkward at best.
Possibly. Now I want to go experiment with XenServer and CEPH.
-
@travisdh1 said:
@StrongBad said:
I am pretty sure that CEPH and Xen can coexist, but I don't know about with XenServer. Doing so would likely be very awkward at best.
Possibly. Now I want to go experiment with XenServer and CEPH.
Would make for a fun project.
-
@travisdh1 said:
@StrongBad said:
I am pretty sure that CEPH and Xen can coexist, but I don't know about with XenServer. Doing so would likely be very awkward at best.
Possibly. Now I want to go experiment with XenServer and CEPH.
The most recent articles that I can find about it talk about the ability to make XS a Ceph Client, but not necessarily a Ceph node. This is the direction I'd like to go long term with our storage situation. Get three whitebox servers with a lot of storage (relative to how much I have had) and run Ceph on them to present a back end for XS.
-
What kind of workload do you run? Mostly Linux, Windows, etc? You have four nodes today, right? Anything keeping you from dropping to fewer?
-
@scottalanmiller said:
@travisdh1 said:
@StrongBad said:
I am pretty sure that CEPH and Xen can coexist, but I don't know about with XenServer. Doing so would likely be very awkward at best.
Possibly. Now I want to go experiment with XenServer and CEPH.
Would make for a fun project.
After I start the next upload I'll have some time. Very experimental as I think I'm going to fire up a XenServer instance in VirtualBox with like 10 10GB HDD, see where I end up and if I finish it before the upload completes.
-
@scottalanmiller said:
What kind of workload do you run? Mostly Linux, Windows, etc? You have four nodes today, right? Anything keeping you from dropping to fewer?
On these hosts it is all Linux. It is mostly processor and memory intensive compute processes with not a lot of storage required at this point. I'm shooting to start out with just two hosts initially and see if, with better management and transparency, we can manage with the two newer hosts and leave the other two for testing or other duties.
-
If you can get down to two, then you can go for bigger hosts down the road. Start with two socket hosts if you want, but you can go to four socket hosts to get double the density without getting more nodes. This allows you to do more and more to stay with less management. Going to CEPH only makes sense if you are going to a lot of nodes. It's worth a lot to go to fewer. Since Linux has no licensing complications from having lots of CPUs like Windows does, you get that extra benefit for "free".
-
@scottalanmiller said:
If you can get down to two, then you can go for bigger hosts down the road. Start with two socket hosts if you want, but you can go to four socket hosts to get double the density without getting more nodes. This allows you to do more and more to stay with less management. Going to CEPH only makes sense if you are going to a lot of nodes. It's worth a lot to go to fewer. Since Linux has no licensing complications from having lots of CPUs like Windows does, you get that extra benefit for "free".
They are already quad socket motherboards, so I have that going for me...
At this point I have zero visibility into what our actual workloads are because of the version of OpenStack Cloud we're running on, so I'm going in a bit blind. That is why I'm going to try to just run two hosts and add as necessary.
My reasoning for looking at Ceph in the long run is that I'd like to centralize all of our storage. We currently have an Oracle (Sun) NAS that is very expensive to maintain, and is a single point of failure. This is where all of our critical data is stored (not my design). It is also the backend for some of the VMs running in our existing cloud.
-
Wow, well that could make more sense then. If the CEPH is going to power a bunch of stuff. Look at Dell R720xd nodes for CEPH, should be pretty awesome.
-
How bad is the OpenStack situation? I mean, would tackling the NAS replacement first, by putting in CEPH nodes, and getting it working as a NAS replacement, phasing that out and having a testing CEPH platform as a starting point and then adding XenServer as needed make sense?
-
@scottalanmiller said:
How bad is the OpenStack situation? I mean, would tackling the NAS replacement first, by putting in CEPH nodes, and getting it working as a NAS replacement, phasing that out and having a testing CEPH platform as a starting point and then adding XenServer as needed make sense?
It is broken at the moment. We're running at about 60-70% of capacity due to one of the servers deciding to take its processors and go home. I'd prefer to do it that way, but my priority is to get everything functional on the front end.
We have decent backups of the NAS and an expensive support contract so downtime there isn't as concerning in the short term.
-
Okay, makes sense then. I'd lean towards XenServer with local storage then. You can migrate to CEPH when it is ready. This would get a single node up "instantly" and let you move others as the opportunity arises.