FIle server help - to VM or not to VM

[REALM]

I need to replace my Win2003 file server which is an old physical DL380 box using san storage for all the data, about 1.5 TB. I also want to split it into 3 different servers to reduce risk and load.

The question is should I:

A) Buy 3 new physical Proliant servers and use san storage?

B) Buy 3 new physical Proliant servers and use local storage since each will only have about 500GB of data.

C) Create 3 new VM's as file servers and don't keep them physical.

I've heard file servers, database, and backup servers should remain physical and don't have many advantages of putting them into vmware.

Is this true? Thoughts? Thank you.

[weinstein5]

How many users are you talking about? I am inclined to virtualiizae all three - Unless all the users are movin GB files constantly I thinl you will find it will work -

[COS · Accepted solution]

I concur.

Buy a single Dual Quad Core CPU DL380G5 with redundant fibre HBA's and a quad port NIC (Or multiple physical NIC's) and use a good fibre SAN. Config RAID for best performance. You really don't need three new physical servers. Isolate each VM to their own NIC. You'll see the same performance because file servers are not CPU or Memory intensive.

But if you need the excuse to get more hardware, sure, say you need three DL380G5's.

:op

[Rumple]

It all depends on what your SAN is and the load on your server. If you have a netapp, just use the CIFS for your file server and call it done.

Personally if you are into that type of physical design you are probably looking at using DFS as you do not want to have to map users to 3 different servers just to get their data. With 3 phycial servers and SAN you've spread out the risk, but a failure drops a 3rd of your business for the time it takes you to recover from the failure.

If you have emc, hp, etc then I would probably virtualize (assuming its not massive load and you don't have a whack of shares). Virtualization at least gives you the ability to build an HA system without dealing with MSCS which personally I find to be a bigger pain then its worth.

[malaysiavm]

I will suggest u virtualize it. Choose RDM for your storage setting in the VM to ensure you have the peak perfomance on your VM.

Malaysia VMware Communities -

[REALM]

Thank you for the help... I'm inclined to go with VM's but our biggest worry is the network I/O as we're looking at about 500-600 users, with maybe 400 using it concurrently.

I have about 20 drive shares on the server, and a total of 1.5TB, probably going to grow by half TB / year.

I have (3) new DL580G5's running ESX 3.5, with 2 nics for console, 2 nics for vmotion, and 2 nics for public per box. Is that sufficient enough for that type of I/O?

[Rumple]

I would probably be inclined to actually utilize a netapp with cifs shares in this case...then its easier for DR solutions (snap to DR netapp and change dns pointer) and lots of utilization available on netapps.

I would drop windows as a frontend completely...

[COS]

I concur!

You can also pool your NICS (I recommend a quad port) into one large aggregate, thus a 4Gb connection. Of course you'll need the network folks involved to set the "trunking" at the switch.

[max_inglis]

If you don't have access to a netapp filer, make sure you use RDM for the disks - vmfs adds a lot of file system overhead, and you'll get better performance from the disks.

That being said, i agree the best solution is the CIFS shares. Just make sure you have the bandwidth available to the guests.

[malaysiavm]

Your bottle neck as you mentioned is related about your network. You can increase to quad port for gigabit NICs, but remember, this doesn;'t really solve your problem. In DC architecture, each of the switches you connected did direct impact the performance of the data flow in your DC. You need to verify your uplink and down stream connection from each of you physical switches you have in DC to your core switches in the DC LAN. If you uplink only 2gb per switches, no matter you team for 10 NIC, it will still not go more than 2gb for the traffic to route from VM to the servers you try to communicate.

http://malaysiavm.com/blog/virtualization-network-storage-data-center/#more-81

NIC teaming will help, and suggest to connect to different switches in the DC.

In that, you also may need to look at the subnet you configure, the less hope you do from your client to the VM, the traffic will go through faster.

Malaysia VMware Communities -