Jump to content

Recommended Posts

Posted (edited)

Scenario (not education)

 

1000 users

6-7+TB of storage

 

requirement - 99.9% uptime loads to be balanced

 

My current thoughts would be to have probably 3/4 servers (in my clusters I always work off the same theory as Raid 5, i.e 3 to accommodation for 1 failure) with data replicated between the three or 4 using DFS (ideally in two seperate physical locations). Ideally I don't want to be using a SAN for the data.

 

My problem is then how to load balance between the 3 or 4 hosts. I know DFS does a limited amount of load balancing but it doesn't take into account current load on each host.

 

I've done some googling around but cannot seem to find any sort of answers to a proper load balancing solution for file servers, web, exchange etc is all fine!

 

Another idea I have had is to use the 4th server (lesser specced and sata drives) as a copy only and then back this up, i.e backups can be run throughout the day without the users noticing.

 

Ideas?

Edited by glennda
Posted

DFS will slow things down, especially if you have large files such as videos. DFS copies each changed in order so if one large file is being syncd, everying else has to wait.

 

I would have to recommend the SAN approach with hyper v. Virtualise the file server which will give you HA. If you use 2012 hyper v, you can do replicas to an offline VM. Any issues with OS then boot the replica

Posted
DFS will slow things down, especially if you have large files such as videos. DFS copies each changed in order so if one large file is being syncd, everying else has to wait.

 

I would have to recommend the SAN approach with hyper v. Virtualise the file server which will give you HA. If you use 2012 hyper v, you can do replicas to an offline VM. Any issues with OS then boot the replica

 

Thought about that but don't really like the idea of keeping all the data on the SAN if i am honest - plus to many connections to keep it all on one host.

Posted

One geographical location?

 

How about hp p4500s or similar, which does enclosure level mirroring. Connected via fibre to three physical servers which then sit behind a pair of hardware loadbalancer?

 

Boom, multi path resiliency and load balancing

Posted (edited)

Don't know if 2012 has improved things any but my experience of replicated 2008 DFS is it is useless for shared folders. It doesn't replicate changes/locked files quick enough so you can have multiple users hitting the same file on multiple servers, leading to inconsistencies across servers. As said everything waits in the que so depending on load it can be minutes before changes get replicated. That was with just 2 servers. I'd hate to try it with 3 or 4.

 

Also you only get a performance boast on read operations as any write request gets replicated across the servers.

 

 

Is there any way to have a failover set up using DFS? You could have 2 beefy servers loaded with SSDs to give you the performance you need, with only 1 getting hit for files thus no replication issues. You could even share the load between the 2 servers by splitting the shares.

 

TBH I think what you really want is a clustered FS, which DFS is not.

Edited by Guest
Posted
Using a file server cluster on Svr 2008 R2 here...two servers, witness and data on SAN. Works well, as can do maintenance on hosts without impact. Only thing Ive not got working is Previous Version VSS - which was being a pain last time I looked at it.
Posted

Shared Storage and Virtualisation and use this to host:

 

Intel® Modular Server

An Awesome box. (Viglen Storage Group sell them. VSG are *really* good at Storage and VM's) but might be EoL now/soon.

 

See this for scalability of Windows Server

FSCT test results detail the performance of Windows Server 2008 R2 File Server configurations - 23,000 users with 192 spindles - Jose Barreto's Blog - Site Home - TechNet Blogs

 

My fileserver is similar in capacity to the one you are building. The underlying LUN is across 10 SATA disks. It is pretty responsive - unless there is some snapshot work going on. Typically at beginning/end of lessons I see 120Mbytes/Sec through the NICs. My bottleneck being the 1GB/sec iSCSI SAN. (why oh why didn't I take the IMS? It has up to six times the bandwidth to the disks for a given VM). I have deployed the latest Kernel NTFS and TCPIP and SMB2 hotfixes to get there though. (As linked from here:List of currently available hotfixes for the Offline Files Feature in Windows 7 SP1 - Romania Networking Team Blog - Site Home - TechNet Blogs)

 

 

Then there is this feature in 2012 which might be worth looking at.

Scale-Out File Server for Application Data Overview

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...