You thoughts on a SAN.

[budgester]

Hi,

 

We are running out of disk space seriously quickly and are looking at implementing a SAN, that should have the capability of being used as virtualisation storage in the future.

 

I currently have a 5406zl core switch and was looking at implementing it in the following way.

 

HP 8-port 10GBASE-T v2 zl Module (J9546A)

 

will give me 8 10GBe ports.

 

And a

 

https://secure.dnuk.com/systems/configure/rs3000x.php

 

With 4 10GBe RJ45 Ports

 

Or should I really be looking at SFP+ connections between the storage and the servers.

 

Comments/suggestions please.

[gshaw]

Personally I'd only go for a SAN from one of the more established brands just from a long-term support perspective, others may disagree

[budgester]

Personally I'd only go for a SAN from one of the more established brands just from a long-term support perspective, others may disagree

 

Such as ... SUN, EMC, HP ????

 

Any other to consider ?

[Abaddon]

We use NetApp.. extraordinary service from them from a support side. Not that we really ever need it.

[gshaw]

Such as ... SUN, EMC, HP ????

 

Won't go far wrong with any of those or the Netapp suggested below, might cost a bit more though depending on how much educational discount you can squeeze out of them

[Oaktech]

we are looking at the EMC symmetrix vmax se, or the clariion CX4 backed up to a celerra nas box in a different area of the site. we are expecting the projects to be costly!

 

although the offerings from huawei seem to be gaining rep and popularity.

[rh91uk]

After going on the EMC course which was excellent .............. I'd reccomend EMC

[Oaktech]

After going on the EMC course which was excellent .............. I'd reccomend EMC

 

just out of interest, what solution are you using from EMC?

[budgester]

We use NetApp.. extraordinary service from them from a support side. Not that we really ever need it.

 

Which model NetApp do you use and how much storage have you got in your array ?

[nicholab]

Don't forget Dot Hill they do some very competitive product and have a licencing mode more akin to Oracle storage products.

[pantscat]

We use Netapp here - a basic FAS-2020 (with an expansion shelf for more disks) - it's great.

 

On the cheaper side of things - I hear Thecus make some perfectly serviceable SANs for smaller environments.

[Duke]

I've had bad experiences with NetApp, but them, EMC and Oracle/Sun would be the bigger names.

[Abaddon]

We have FAS 2020's and FAS 2040's here. Total storage capacity is 65 Tb all told, spread between the filers and additional shelves on some of the heads. Edited November 25, 2011 by Abaddon

[zag]

Sans are a bit overkill for me in a school.

 

We run 2 x normal Raid5 storage servers which run the same data (1 is just a reserve if anything goes wrong) and it works perfectly.

 

We virtualize here but I really don't see the advantages of of storing the data on a SAN. Seems like its just lining up a single point of failure.

 

I can understand it if I was running 100 virtual machines in a data center or something like that but we run less than 4 on each hyper-v host so local storage is fine for that.

[RabbieBurns]

we went with HDS for our SAN, theyve been great and have a local presence which is handy..

 

@zag, im at about 46 VMs, just now.. the SAN has 2 controllers, 2 switches, and 2x 2x10gb into the core, and each of these are spread over 2x UPS. As well as all the raid groups in the san there are also hot spares etc. I dont see where the single point of failure comes in?

[Abaddon]

Not to mention the iops available on a dedicated SAN solution, which is way in excess of a storage server. It's about speed as well as resilience.

[Oaktech]

@zag a good quality, well implemented SAN is about as far from single points of failure as you can get, multiple drives with constant data shadowing and 10minute backup to disc schedules, multiple power supplies with UPS's (powered from separate circuits if not separate supplies if done correctly) multiple separate spine connections within the appliance, automatic failovers, multiple redundant connections to physical hardware... the list goes on!

[cpjitservices]

may I recommend Pillar!

 

Oracle and Pillar

 

Oracle has just bought them

[zag]

46 VMs!!! wow yeh you are getting into datacenter territory there I have 16 here and thought that was a lot!!

 

Basically what I'm asking is what if the SAN fails? Does it just auto repair itself like a RAID5 machine? Is it more reliable than a normal computer?

[Duke]

Basically what I'm asking is what if the SAN fails? Does it just auto repair itself like a RAID5 machine? Is it more reliable than a normal computer?

 

In an ideal world... you have two (possibly more) separate SANs, ideally in separate locations, and both of these SANs have dual redundant controllers/heads. The SANs will replicate data between them on a synchronous or asynchronous basis. The disk trays on the SANs will be running at least RAID5, more like RAID6 or RAID-TP (depending on their workload, you might have RAID10 for performance - my point is that there will be plenty of redundancy for failed disks) and there will be hot-spare drives available. Each controller will have multiple NICs with multihoming set up on them. The controllers and disk trays all have redundant hardware - dual PSUs, dual SAS controllers/cables, etc.

 

If a disk fails - it'll rebuild the array using a hot-spare until the drive is replaced. RAID-TP would take three failures before you had any concerns.

If a controller fails - it will failover to the redundant controller with zero (or very, very little) downtime.

If a switch/link fails - multihoming or port trunking will ensure a route to the SAN is always available.

If lots of disks fail, or you suffer a catastrophic failure to the SAN - you will automatically or manually failover to the replicated SAN with very little data lost (depending on your replication schedule).

If data is lost accidentally - your SAN usually has snapshots which can be treated as an internal backup and will allow you to get data back if your normal backup strategy fails.

 

In other words, if done properly you'll have no single point of failure. The above sounds complicated, but if you buy enterprise-level hardware then it should (not arguing with those using standard servers and rsync!) be simpler and more robust in the event of a major failure than trying to do the same with Windows servers or some kind of local storage.

 

Chris

[Finch7]

we went with HDS for our SAN, theyve been great and have a local presence which is handy..

 

@zag, im at about 46 VMs, just now.. the SAN has 2 controllers, 2 switches, and 2x 2x10gb into the core, and each of these are spread over 2x UPS. As well as all the raid groups in the san there are also hot spares etc. I dont see where the single point of failure comes in?

 

Corrupt firmware!

 

Of course only if you mess around updating firmware.

[Duke]

Corrupt firmware!

 

 

That's what the dual controllers/heads are for in theory. You should be able to update one, when you reboot it for the update to take effect it'll failover to the other head, you can then check the first head is running properly before bringing services back on to it, and if anything breaks when you do so you go back to the second head that's still on the original firmware. Like I said, in theory!

[gshaw]

Depending on your SAN the controllers load balance and failover in different ways... our VNXe is active-active so I assign the datastores alternately between controllers... when we did a shutdown recently it failed over seamlessly to the other controller and back again.

 

Out of that list the only part I don't have is a 2nd SAN (or 2nd building to house it in actually )

 

Veeam does help there though as I then have a secondary copy of all the VMs that can run directly from backup so that helps. The only thing I'm not entirely convinced about on SANs is using file shares on them natively. When we got the VNXe I was quite excited by that feature as I never wanted to see a Windows file server ever again but unless you use the SAN replication backup becomes more trouble than it's worth so going to just stick with Windows VMs and Veeam as per the rest of the servers.

 

Read an interesting point on PC Pro saying don't virtualise file servers but if the storage is on the SAN it makes little sense not to virtualise and take advantage of multiple host failover and not be reliant on physical server hardware for that role?

[Duke]

Read an interesting point on PC Pro saying don't virtualise file servers but if the storage is on the SAN it makes little sense not to virtualise and take advantage of multiple host failover and not be reliant on physical server hardware for that role?

 

I can only assume it's because you don't want the Windows overhead, plus the hypervisor overhead, then the SAN OS overhead when it comes to serving large amounts of files. That seems to be a lot of IP stacks to go through compared to sharing files straight from the SAN via SMB/CIFS. (definitely my preferred option when possible!)