What do I need for HA and VMWare?

[Steveocee]

Hello,

I can't seem to put my finger on the right information so I'll explain what's happening and see if anyone can shed some light on it for me.

Currently we have 4 HP DL380G7's all machines are equally specced.

We have machine 1 running ESXi free for some "non-essential" services on VM's as a test of ESXi and to see how we got on with the software with the hope of machine 2 being a failover (then found esxi can't do this)

We have machine number 3 also doing similar, running some "other" services with a free version of ESXi and machine number 4 powered off as it was going to be used as a failover also.

My "grand" idea was to pool all 4 machines into 1 resource, I'm fairly sure this is possible from another thread I've seen but am wondering exactly what I'd need to propose to do this?

I appreciate we'd need some VMWare licenses to do this and this could work out quite expensive as each machine is dual processor (2 hex core's in each).
Also what about the HA option? Do they need to have some kind of networked storage or can the storage of the machines themselves be used?

Please throw some ideas at me.

 

[TrojanWhore]

hi Steve,

I can't help much with the technicalities I'm afraid, but it is definitely possible to configure clusters of multiple hosts, a place I used to work had one cluster of 3 very high spec servers, with a cluster of a few older machines in another building to manage connections and replication between the 2 server rooms as well as failover exchange and DC.

In terms of storage, we ran them on thin provisioned shares from our SANs, the main reasons for this were:

- easier to manage volume sizes without taking anything offline
- snapshot based backups so you can restore to a point in time etc.

so I think the question on internal storage comes down to what the alternatives are - If you have a SAN with space to spare, go with that, if your current set up is a NAS or similar, then I'd probably go with internal disks in RAID1, and replicate the important volumes overnight to the NAS or other backup solution.

*disclaimer - whilst I know little bits about infrastructure stuff, I've mostly just worked as a helpdesk technician!

 

[Steveocee]

If we needed to buy something SAN and it is a justifiable advantage then there is no reason why we wouldn't do that. This will ultimately be running some very dependant production systems on if it is applied correctly.

 

[nutcase]

Stick to replicated local storage, not a SAN. If you buy a SAN, you then need a second one or it's not HA...

 

[si_]

Hi,

ESXi free is good but has limited functionality and without vCenter/shared storage you won't be able to use some of that like HA, DRS etc.

Local storage can be used but when a host fails so does the local storage and therefore the vm's running on it also fail (unless you are using in guest clustering I guess). vSphere would HA restart those vm's in around 3-5 minutes if on shared storage. If the host is toast (that rhymes!) then conceivably the disks are dead and you will need to copy the vm back from wherever you copied it too - depending on many things that could take a while and may not be what is acceptable to the business.

vSphere Essentials Plus kit for 3 hosts would probably be useful with some form of shared storage between the hosts.

SAN is generally going to be more HA than internal storage on a host unless you go vSAN.

Also the switching infrastruture between hosts will need to be HA i.e. Fibre Channel or 10Gb/1Gb network switches.

Also need to check out backups for the vm's then - free ESXi does not have the data protection API's available so you would have to use in guest backups which is a pain. vSphere Essentials Plus has an included backup option I beleive. Veeam do a small backup essentials product I think.

Si

 

[Sparky__H]

Also look at Simplivity, the omni cube is great but if you want a pair in HA your going to need a directors sign off...

maybe a hand shuffle too...

 

[Vince]

I have 4 dl380 g7's running ESXi over two sites running HA and DRS with a P6000 6550 EVA Fiberchanel back end so can probably help since my servers are identical to yours apart from ive upgraded each of mine to 192gb of ram the servers also run fiber networking.

Firstly you will need to license your servers. Now I have two separate licenses that cover 3 servers each so two separate vcenter deployments and this is cheaper than trying to license an extra server into the first deployment even this way your looking 15k ish (might be less ill have to check) pa licensing. Put simply licensing is horrendous above 3 hosts so if you can spec up 3 and get away with it then that's the sweet spot (essentials plus as pointed out above). When you have a vcenter deployment you will be able to manage all your hosts from a single point utilizing vcenter. Like my screenshot below:



You will need to consider storage and HA and DRS, vMotion will be a problem without supported back end infrastructure, really you are going to need a dual controller san, fiber switches, and a stack of disks. I run 5 shelves in total all stacked with 600gb 15k sas. Something like EVA, 3par or Pure are perfect but if you worried about licensing vcenter then affordability on this kind of back end might be an issue.

There are of course considerations in config on the management network for HA and a host of other things including backup (veeam is awesome) but you have much more to do before this so small config details really are way down the list.

Stick to replicated local storage, not a SAN. If you buy a SAN, you then need a second one or it's not HA...

This is not strictly true and many storage solutions can suffer massive disk failures without even so much as missing a beat even losing a whole tray is possible while remaining running. Many sans don't run traditional raid configs in terms of where data is stored. HP's EVA implementation uses redundant storage sets, with vraid on top, at the business end so long as you aren't massively over provisioned you can suffer multiple simultaneous disk failures while not impacting availability.

The 3par takes this a step further and uses chunklets and each drive is chopped into 1gb sections where they apply some crazy algorithms. effectively applied raid over small storage sections of the drive with all drives in the set used for parity.

 

[Caged]

Stick to replicated local storage, not a SAN. If you buy a SAN, you then need a second one or it's not HA...

A SAN is highly available by default as it will have dual controllers, dual paths to disks, and redundant links to servers.

This seems like a disproportionate of money to spend on fairly old servers, though. Rather than investing in creating shared storage via a physical or virtual SAN, have a look at deploying HCI with something like Nutanix, NetApp, Cisco, or a qualified Windows Server Hyper-Converged design.

 

[nutcase]

Sorry, I should elaborate a little bit - the environment the OP is talking about is too small to consider a SAN. Replicated local storage (Hyper-convergence I think is the current buzz word for it) is ideal for that size. A SAN is not just the disk storage - you're talking about dual SAN switches etc. as well - https://en.wikipedia.org/wiki/High-availability_cluster

 

[Steveocee]

For the purpose of this. Lets assume all 4 servers are in the same rack with access to the same management network.

 

[blueboy2001]

What are your recovery time objective and recovery point objective?

Recovery time objective (RTO) is how long from a failure you want service restored and applications back up and running. Recovery Point Objective (RPO) is how far back you’re willing to restore to - essentially much data you’re willing to lose.

VMWare HA only really protects you against host failure, but with an RTO of a few minutes and an RPO of more or less immediate at the point of the host failing if your application is reasonably tolerant. This requires shared storage in some form, usually a SAN. An entry level dual controller SAN like a HPE MSA 2050 can be had well under £10k depending on disk config. You can run these using iSCSI at 1Gbps but the unit supports 10GbE natively so you can upgrade later - eg. HPE do a 12port 10GbE 1950 switch with 4SPF+ ports, use a couple of DAC cables to connect the SAN and some 10Gbps NIC’s in your servers and you’ve got a decent infrastructure for not a huge outlay. I’d get a VMWare Essentials Plus Kit which is £2.5k PA and build a 3 node cluster. The MSA’s are a good solution because they scale fairly well and the support doesn’t go to crazy money after 3 years like some vendors.

A cheaper option would be to get a 2x VMWare Essentials Kits at about £600 each and 4 sockets of Veeam Essentials at £1600 total and use Veeam to replicate your VM’s to a second host. This does not require a SAN - you can use the existing local storage. You don’t get HA, but Veeam will allow you to do planned fail overs or quick migrations to your secondary host - these are not as seem less or automated as VMWare HA or vMotion, but can be achieved at much lower cost. If you buy Veeam you’ve also got arguably the best VMWare backup solution on the market as well. You only need to licence the hosts your backing up or replicating from, not the standby hosts with Veeam. All hosts will need VMWare licences. The licence costs I’ve quoted are perpetual, there will be annual maintenance of around 20% after the first year, and they are list prices so you may get a bit of discount off those figures from a friendly reseller.

To give you the best solution, I’d want to understand a bit more about the business and the applications. Solutions should be driven by the requirements of the business and not what the IT department think is best.

 

[Caged]

Sorry, I should elaborate a little bit - the environment the OP is talking about is too small to consider a SAN. Replicated local storage (Hyper-convergence I think is the current buzz word for it) is ideal for that size. A SAN is not just the disk storage - you're talking about dual SAN switches etc. as well - https://en.wikipedia.org/wiki/High-availability_cluster

iSCSI is dirt cheap though. The HP MSA series classes as a SAN, and with a couple of switches with decent packet buffers is perfectly capable. I would agree that it's pointless spending that much cash on something that is just storage though when it's the year 2017 and way more options exist.

 

[Hellsmk2]

Stick to replicated local storage, not a SAN. If you buy a SAN, you then need a second one or it's not HA...

Eh?

For HA you will need shared storage as mentioned... a SAN will do the job, or vSAN if you don't want another device to manage.

You will then need to deploy a virtual centre and create a data centre and cluster (2 min job... just google it ), and then add your esx hosts.

At that point you can enable HA on the cluster (30 seconds). Job done.

 

[Hellsmk2]

snip

Sorry but I think you've completely missed the point. He's not asking for a DR solution, and certainly not one where he's replicating to an esxi server in the same room. That's just a massive waste of resource and pretty much a bad idea.

.... also if it was replication he was after, you could use vsphere replication for free.

For a HA cluster it's fairly simple. Some form of shared storage. A virtual centre. Appropriate vmware licensing. That's it.

 

[ecksmen]

Storage Array != SAN. Once you have decided on the hardware do not forget that to get the most out of the solution you'll need licensing to gain desired functionality. For example VMware licensing you may decide to take advantage of Enterprise Plus for distributed network capability (not required, but desirable) and potentially additional licensing for applications and hope to god it's not Oracle due to being able to move (ie DRS or (s)vmotion) that may conflict with current licensing arrangements.

 

[Steveocee]

Hi everyone,

So I've harvested all of the ideas posed and off the back of it have approached a local VMware advertised partner to come in and advise us what they think we need against what we think we want and then critiqued against the advice you guys have given.

It looks like shared storage is the preferred and then using the machines around it clusteres with an essentials license. Although that won't get the HA feature we want it looks to be the first step in the right direction. Once we have the platform up and running we would then look to upgrade to the plus kit where it includes the HA features and then back it up to something else with Veem.

Thank you all for your advice.

 

[Caged]

Have I missed where you've said what these servers are actually doing?

 

[BongoHunter]

I would think with servers of that vintage and the cost of changes needed to get what you want in terms of availability you could just go with a pair of much more recent boxes with local storage and 10g+ ethernet. If running essential stuff - Hardware on support can be a lifesaver