Cannot decide between 1U servers for growing company
-
as I think about it, the LUN disconnect / reconnect might work as it is Linux handling the connection on both ends. Assuming you have Xen with Linux, but that's a safe assumption.
-
We use lots of 1U Servers.
Only servers that are 2U are backup appliances (which have 40TB of storage)
However, that being Said 400 is relatively small and when I worked at companies that small we usually used 2U with local replicated storage.
Also do not used ProxMox.. Just don't it's a toy.
-
@coliver said:
@scottalanmiller said:
@coliver said:
@ntoxicator said:
So if anyone can explain to me
To do away with centralized storage such as what we have now and I've been moving to. I suppose this is what I've grown use to.
In order to have localized storage at the node/hypervisor level. one or many of the hypervisors would be storing all the data and sharing out the NFS? Then its replicated between? probably with DRBD Storage.
however, would would this be done with Citrix Xen Server for instance?
You could also look at Starwinds Virtual SAN. Which could do this as well.
But is limited to Hyper-V for best results and can make due in the VMware world but is inferior (but you shouldn't be looking at ESXi anyway so not a big deal.)
Doesn't work well with Xen? I thought they supported it.
If it can be used with Xen, which I am unaware of being supported or available, it would only be able to do so in the "Vmware" style fallback VM mode which is vastly inferior to DRBD. It's their integration with Hyper-V in the Dom0 that makes them super powerful there. Definitely not happening on Xen today.
-
Ok so I'll scratch the Proxmox ideal. As right, it probably wouldnt scale. meh.
But I do know and have been planning to migrate away from iSCSI over to NFS storage for Xen Server. I already started... but the issue is with migrating the storage from the current local disk to new local disk that gets assigned to the VM
Still 100% confused on local replicated storage.
Scott made a comment on Xen Server with HA-Lizard
But wouldnt all that storage replication STILL be handled over 1Gbe backbone??!
-
@ntoxicator said:
Scott made a comment on Xen Server with HA-Lizard
But wouldnt all that storage replication STILL be handled over 1Gbe backbone??!
Yes, ONLY way to avoid that is to abandon HA and move on. HA requires certain things that you can't get away from even if you improve the architecture.
Our Scale cluster has a dedicated 10GigE SAN network, for example, to overcome that. Huge throughput and sharing nothing with other functions.
-
UPDATE
Oracle just got back to me on Pricing. Made me puke
10K PER server for a F*** simple 1U box? what f[moderated] is going on with this market, have I completely lost touch?
-
@scottalanmiller said:
@ntoxicator said:
Scott made a comment on Xen Server with HA-Lizard
But wouldnt all that storage replication STILL be handled over 1Gbe backbone??!
Yes, ONLY way to avoid that is to abandon HA and move on. HA requires certain things that you can't get away from even if you improve the architecture.
Our Scale cluster has a dedicated 10GigE SAN network, for example, to overcome that. Huge throughput and sharing nothing with other functions.
It doesn't have to. @ntoxicator was talking about installing a 10 GB network for replication.
-
@ntoxicator said:
Still 100% confused on local replicated storage.
Which aspects? Think of it like SAN. If you want HA SAN you need a replicated cluster of SAN. If you do that, the replication handles one point of fragility but you still need to buy multiple SAN devices, have a network to connect them to the compute nodes, etc. If you move from SAN to DAS you can eliminate some of the networking complexity meaning you get even safer and less complicated, but you still have extra devices to fail and more cost.
Take this to the next level, move the DAS from external to internal and have it be part of the compute devices. You lower cost by removing the extra boxes and reduce risk by reducing the number of parts.
Each move towards local increases reliability and lowers latency and increases bandwidth while lowering cost. It's all win.
-
Right. Future plans was to do 10Gbe backbone for ONLY the Hypervisor Nodes & the NAS/SAN
All other network traffic (workstations, phones) - would be handled by the 1GigE network switching
I would just inter-connect the 1GigE network to the 10GigE network to be able to talk and have access to those devices.
Ofcourse I know local storage would be faster.
Just looking for cost-effective.
As you know I was looking at 2X Synology rackmount 12-disk units (replication - HA setup). and then just Hypervisor nodes.
But damn.. Oracle hit me with 10k per 1U server. what the hell
-
@ntoxicator said:
UPDATE
Oracle just got back to me on Pricing. Made me puke
10K PER server for a F*** simple 1U box? what f[moderated] is going on with this market, have I completely lost touch?
It's that you have moved into a new Oracle hardware world. They don't make servers for generic consumption like this any more. They still make great stuff. But they are focused on vertical integration for Oracle workloads. I don't look at them for AMD64 platform stuff anymore. It's Sparc only now, and I never get to work with people wanting that size of RISC gear these days.
I've got a "new" big Sparc RISC server slated for our lab, though. Hoping to get that bought and racked in 2016. Large scale Solaris system.
-
@ntoxicator said:
But wouldnt all that storage replication STILL be handled over 1Gbe backbone??!
Dual port 10Gbit ethernet cards cost as low as $400-$500.
-
@ntoxicator said:
UPDATE
Oracle just got back to me on Pricing. Made me puke
10K PER server for a F*** simple 1U box? what f[moderated] is going on with this market, have I completely lost touch?
Check out http://www.xbyte.com/.
-
@ntoxicator said:
As you know I was looking at 2X Synology rackmount 12-disk units (replication - HA setup). and then just Hypervisor nodes.
But damn.. Oracle hit me with 10k per 1U server. what the hell
That's four or five total nodes. How much total storage needs?
-
Right, which I was prospecting to have 3 large Hypervisor nodes to handle workload.. but might need to scale larger to handle the future
again, company plans to have 400-500 employee's by year 2020.
Total storage. I would guesstimate would grow larger than 5TB in 2+ years.
-
@ntoxicator said:
Total storage. I would guesstimate would grow larger than 5TB in 2+ years.
That size is easy to handle. No problem there.
-
@ntoxicator said:
Right, which I was prospecting to have 3 large Hypervisor nodes to handle workload.. but might need to scale larger to handle the future
again, company plans to have 400-500 employee's by year 2020.
Total storage. I would guesstimate would grow larger than 5TB in 2+ years.
That really isn't that much. You could look at a middle of the road R720xd and get several times that amount in just one server.
-
But all these 400 future employee's using a RDP wrapper to launch their software? Similar to thin-client. (2X Application Server)
I do not install the needed customer software on the employee's workstations. Much easier to handle. I'd be putting out even more fires on daily basis.
-
its Billing Related software.. medical billing. as FYI
So we deal with numerous different EHR/PM systems. One day looking to be able to streamline to one central software. This is a challenge
-
@ntoxicator said:
Right, which I was prospecting to have 3 large Hypervisor nodes to handle workload.. but might need to scale larger to handle the future
again, company plans to have 400-500 employee's by year 2020.
Total storage. I would guesstimate would grow larger than 5TB in 2+ years.
That's Pretty tiny really.
-
@ntoxicator said:
again, company plans to have 400-500 employee's by year 2020.
Long term growth is actually a spot where hyperconverged solutions will normally shine. They have much better growth paths in most cases. With the Synology approach, for example, you will be at your maximum storage performance (your main bottleneck) on day one and as you grow you will have no good performance growth path outside of ripping and replacing. Going with something like Scale or similar HC paths you grow simply by adding a node. So you could start "small" today meeting today's needs and only buy more as needed in the future so only investing when additional capacity is needed which both puts off the purchase to save money (time-value of money) and reduces risk by only buying capacity when it is needed, not buying it and risking it never being needed. The HC approach grows your storage (capacity AND performance) while you grow compute in lock step so you don't get stuck having to rip and replace to upgrade your bottleneck.
