Need more storage for servers - What to buy?!

[PotNoodleTech]

You should probably justify that with schools under X users.

Some large schools and MAT's have in excess of 5000 users.

 

So, talking of justification, why do large establishments require a SAN as compared to Local/NAS/Direct Attached SAS as discussed?

[mjk]

So, talking of justification, why do large establishments require a SAN as compared to Local/NAS/Direct Attached SAS as discussed?

It would depend upon the volume of transactions - a larger establishment typically has more. For example, a school with 2000 external users who rely on a virtual learning environment, or one with 3000 parents who all log in to get parental reports, or one with remote desktop or VDi might have more use for SAN than a primary school with 20 staff.

[richbrowncardiff]

The thing is a SAN is just a term rather than a specific hardware platform. Back in the day Sans were all the rage and very very expensive but if we take the definition of a san it is storage attached to the network and presented typically via iscsi. Which these units all Do very well. its like everything you can spend as much or as little as you like but generally these days a well specified setup will work for thousands of users without any problems at all. BTW not being awkward just wanted to give my personal view for what its worth. In a school we need high availability and reliability but at a small cost this is where the synology / qnap platforms come in as they have a level of performance which is good enough for the setting. i wouldn't use them in a bank but will use them all day in a school, college and university even with 10,000+ users

[bossman]

@mjk

 

Its all about IOPs and connection speeds, for a MAT with 5000 users if you had 10GB backbone and each Server device had the same with a good quality set of NAS units with SSDs and at least a 1GB bandwith internet connection then you should have no problem with VLE or external access.

 

Infrastructure is key then system hardware etc etc

 

No need to fork out vast amounts of money for redundancy and speedy connectivity these days.

[Davit2005]

We did a FibreChannel SAN refresh at the last place £160K and this did not take into account the 2 SAN switches, FC SFP's and Server HBA's we had already.

 

Personally for iCSCI I'd go for 2 separate switches, make sure the SAN has redundant PSU's and controllers for resilience and redundancy. In VMWare you can setup round robin and adjust the IOP's to get the most out of the network connections even if you are only using 1Gb connections.

[pcstru]

For me, when you say SAN what is most important is that the network is separate and only carries storage traffic and hypervisors and servers need multiple NIC's to be able to talk to both the SAN and the application side. In terms of the storage, you can then put cheap iSCSI NAS on that network or a more expensive solution.

 

Personally, I justified ~£30K on two Dell servers with 24 SAS spindles in each running some cheap SAN software (OpenE) to provide a solution in which data is duplicated across the nodes and we have some ability to automatically fail over if there are issues. That was 4 years ago so it may now be that budgets mean we cannot replace that with a system which is as capable, but I will want to find a way to get as close as possible first for duplication of data and second as a hot standby.

 

Our users files are on cheaper iSCSI NAS (various Thecus units but primarily two N12000's, one of which is synced overnight as a hot standby). IMO the price of a second unit is a lot less than the potential cost of not having data available for a week while a new unit turns up and we populate it with data from backups (an activity which will take several days to complete).

[maxrebo]

Ok, so I now hve my Synology as discussed above, I plan to set up an iscsi lun target and add this to my Vmware hosts as a datastore...

First question, I currently have three hosts serving my VM`s ATM on local storage, can you mount this target on all three hosts? (So i can move some of these VM`s from local storage to the Synology box

[Davit2005]

Yes, VMWare does this natively, from memory on the Synology you have to tell it you want more than one host to access the target.

 

Think about setting up an extra switch for redundancy if you are running the OS disks from the Synology.

[maxrebo]

Yes, VMWare does this natively, from memory on the Synology you have to tell it you want more than one host to access the target.

 

Think about setting up an extra switch for redundancy if you are running the OS disks from the Synology.

thanks for the info,

currently this will be attached to one our 2 core switches, will that be sufficient?

[PotNoodleTech]

For me, when you say SAN what is most important is that the network is separate and only carries storage traffic and hypervisors and servers need multiple NIC's to be able to talk to both the SAN and the application side. In terms of the storage, you can then put cheap iSCSI NAS on that network or a more expensive solution.

 

Personally, I justified ~£30K on two Dell servers with 24 SAS spindles in each running some cheap SAN software (OpenE) to provide a solution in which data is duplicated across the nodes and we have some ability to automatically fail over if there are issues. That was 4 years ago so it may now be that budgets mean we cannot replace that with a system which is as capable, but I will want to find a way to get as close as possible first for duplication of data and second as a hot standby.

 

Our users files are on cheaper iSCSI NAS (various Thecus units but primarily two N12000's, one of which is synced overnight as a hot standby). IMO the price of a second unit is a lot less than the potential cost of not having data available for a week while a new unit turns up and we populate it with data from backups (an activity which will take several days to complete).

 

You spent £30k on SAN and don't even store your user files on it? *confused*

[mjk]

You spent £30k on SAN and don't even store your user files on it? *confused*

File storage is near line. Putting user files on your fastest storage system is a waste of resources if you need that for disk access eg for booting your servers.

[PotNoodleTech]

File storage is near line. Putting user files on your fastest storage system is a waste of resources if you need that for disk access eg for booting your servers.

 

There's a couple of reasons I disagree.

 

1. You can never have too fast or too reliable storage for user files. Why put them on slower less reliable hardeware? Do you class your user files as unimportant? Kings college did that too and look where it got them?

2. Being tied into the actual [NAS] hardware is really BAD. IMHO every service you run should be on a VM so it is hardware agnostic and easily moved between hardware. With ESXI and Hyper-V this is a reality. Because my file servers are Hyper V VM's all on SSD storage, If anything happens I can restore the entire VM from Veeam at lightning speed. If something really really desperate has happened. I can even restore them onto a Core i5 workstation with a 6TB drive in it to get the network running in an absolute emergency.

3. You say putting user files on your fastest storage system is a waste of resources and you then say save your hideously expensive san just for booting VMS? I don't know about you but I only boot my VM's once every [xx] days usually whenever they reboot for windows updates. Now if your running multiple SQL servers or Remote Desktop servers that's one thing, but booting?

 

Not having a go at anyone just confused with some decisions made!

[pcstru]

You spent £30k on SAN and don't even store your user files on it? *confused*

OK, not quite true. The SAN is almost exclusively provisioning lumps to our ESX hosts in our cluster (6 hosts with ~30 production servers). Transactional databases (MIS, catering etc) and other systems data effectively live on it as do files in Sharepoint (so there is a small amount of user files on there). However, it would be an expensive way to provision fairly dumb file storage (which is relatively poorly managed by the end users) and we do not IMO need automated failover for that. The performance of the T12000's is fine for our use (~2000 users, ~800 peak concurrent) and if one fails we can switch to the other in a few minutes.

[mjk]

There's a couple of reasons I disagree.

1. You can never have too fast or too reliable storage for user files. Why put them on slower less reliable hardeware? Do you class your user files as unimportant? Kings college did that too and look where it got them?

 

So my take on this is that 99% of files are never access therefore it is a sensible strategic decision to use nearline storage, this is what Google,Amazon and Microsoft suggest for their cloud based systems.

Yes you could copy all of your data files directly into redundant RAM and have super fast access but it isn't necessary.

2. Being tied into the actual [NAS] hardware is really BAD. IMHO every service you run should be on a VM so it is hardware agnostic and easily moved between hardware. With ESXI and Hyper-V this is a reality. Because my file servers are Hyper V VM's all on SSD storage, If anything happens I can restore the entire VM from Veeam at lightning speed. If something really really desperate has happened. I can even restore them onto a Core i5 workstation with a 6TB drive in it to get the network running in an absolute emergency.

 

You are not tied to a NAS supplier, you are tied to the software be it NFS,SMB,iSCSI. You will always be able to restore from a workstation as long as it can read the filesystem and most NAS devices are unix/linux based and 99% sure you will be able to read that on a workstation (maybe your workstation software is limiting this?)

 

3. You say putting user files on your fastest storage system is a waste of resources and you then say save your hideously expensive san just for booting VMS? I don't know about you but I only boot my VM's once every [xx] days usually whenever they reboot for windows updates. Now if your running multiple SQL servers or Remote Desktop servers that's one thing, but booting?

That rather depends upon your IOps and budget

 

Not having a go at anyone just confused with some decisions made!

same

[pcstru]

Not having a go at anyone just confused with some decisions made!

It's an area I find very difficult to get a good balance (between value and risk mitigation), particularly in our environment (schools) where in a revenue business it would be easier to assess the actual costs involved in downtime and data loss, which is a big part of the risk assessment. It's not that I think user files are 'less important', but I think the data loss window we have in the event of complete loss of a single device is reasonable given the costs at the time of reducing that window further (i.e. another £30-40K rather than (ISTR) ~£10K for the two T12000's fully populated with drives). To make sense, a days data loss of user changes would need to involve costs of ~£10K if I'm generous and spread the capital over an equipment lifetime of 3-4 years (all of the kit is now older). The main SAN is addressing a different risk, the total operational loss of systems (in the event of the failure of an entire storage node), which would, if we had no redundancy, probably involve a week or more of major disruption while we provision replacement kit and pull in the backups.

 

Believe it or not, our main SAN was built on the cheap (by virtue of using OpenE) compared to off the shelf options from our previous vendor and other suppliers. I think we got really good value even if OpenE has at times been a pain.

[maxrebo]

Ok, so I now hve my Synology as discussed above, I plan to set up an iscsi lun target and add this to my Vmware hosts as a datastore...

First question, I currently have three hosts serving my VM`s ATM on local storage, can you mount this target on all three hosts? (So i can move some of these VM`s from local storage to the Synology box

 

Just currently setting up my Synology, I have rouhly around 20tb to play with, Do you have to use the entire volume for the ISCSI Lun volume? or could I use some of the space for a Network share?

Its Just i probably dont need all that space for my Datastore(s)

[Paid_Peanuts]

May I ask which Synology did you choose?

 

I am currently considering their FS2017 fully populated with 1TB SSD using an optional pair of 10GB SFP+ cards to connect it to my existing 10GB iSCSI switches to which I have some ageing 10GB iSCSI EqualLogic. Due to the very high replacement price of these EqualLogic I am looking at alternatives and the Synology seem to tick all the box's.

 

Currently we operate 3 hosts running VMware, connected via a pair of 10GB SFP+ to a pair of independent iSCSI switches. The hosts then have 10GB SFP+ connections to my core switch for VM traffic. All iSCSI traffic goes via its own pair of switches for redundancy.

 

I am thinking of adding in a FS2017 and connecting it to each of my iSCSI switches and filling full of SSD to get something quicker and a lot lot cheaper than any tier 1 manufacturer can offer me.

A fully populated FS2017 with 1TB SSD is circa £10k!

 

It's just finding others that have taken the leap to Synology. The last thing I'd want is for it to be temperamental and have to say reboot it once a week/month.

I haven't had a single problem with the EqualLogic devices (of which there are 3) and am always weary about the unknown. Then again I thought that when I moved from Cisco wireless to Ruckus and that is by far a better solution.

[maxrebo]

Im looking at buying another of these to house our MIS server (amongst others). Ive been quoted for WD Red Pro disks @ 7.200RPM (Sata 6gbps), do you think these will be ok for a SQL VM?

[mavhc]

How much data do you need to hold in your SQL database? It's the perfect fit for SSDs usually, just have a mirror of 2 xGB ssds, will be 10 times faster if you don't need TBs of data

[maxrebo]

How much data do you need to hold in your SQL database? It's the perfect fit for SSDs usually, just have a mirror of 2 xGB ssds, will be 10 times faster if you don't need TBs of data

 

Its for our MIS Server, so only 100gb ish. What size SSD`s do you have?

[mavhc]

Could of 256 or 512 if you can afford it. Big question is which model. What's your budget?

[maxrebo]

Could of 256 or 512 if you can afford it. Big question is which model. What's your budget?

 

I've been quoted for 4 x 500gb Samsung Evo 850`s

[mavhc]

Could do it with 2, of course if you have 2 spare use them for your OS on that/another server

[CAWJames]

We had a quote to upgrade our 2 node cluster. I gave our supplier a pitifully small budget to get 2 hosts with hundreds of gigs of RAM, dual Xeons etc, and 20 odd TB of useable storage. I was expecting 2 servers loaded with local drives and some plan to Storage space them or at least some compromise, but, and I still don't see how, they came back with 2 suitable servers, a SAN with 21TB of useable after RAID storage, including a tiering licence and the required SSD's in the SAN to use said licence, plus all the 10GBe cards needed to create our various networks(DMZ's, LAN ,DPM, Replication etc). All this for under £20k! All Lenovo kit. Could have knocked me over with a feather!

[mavhc]

£250 for a WD red pro 8TB, so 6 of those, £1500, £3000 for the servers each with E5 and 2x 10Gb, so 2 of those for servers, one for san, say £1k for SSDs, what was the other £9k for?