File Server Cluster Capacity

All,
I recently started with a new company and I've been put on the cluster
team. We are in the process of redesigning how the cluster works and
we think we have a good design, but not sure how it will scale and how
others have implemented things. Please comment on this with ideas!

Here goes - we have a 4 node cluster that is just used for file
services. The nodes are SAN attached. We have about 7 Datagroups on
the cluster with around 1 TB of data per group. The old way was to
carve 300 to 800 GB disks out of the SAN and mount them as disks in the
cluster. Associate multiple disks with each data group and setup the
file shares, etc. The problems we faced deal with the disk
occassionally doing a chkdsk and taking awhile to come back online.
Users don't like that.

In the new design, we create a 1 GB disk (assign a drive letter) and
then create empty folders for Mount points. We then create 300 GB
disks to be used as mount points. The data goes on the mount point and
the shares point to the data. Also, we have setup Shadow Copy to use
10% of the mount point space for shadow copies. Hopefully the smaller
300 GB mount points will come online quicker if there is a chkdsk
event.

The questions we have are:
1. How many mount points (or how much data) can we/should we add to a
single drive letter? How long will it take to fail over to another
node?

2. How many files shares should we create on a single data group?

3. How much RAM is required? How quickly will the RAM cache be able
to get moved to a new node on fail over?

4. What dependencies do we need to create?

I'm sure there are more questions to be added, but thought this would
start the process/discussion. I've looked all over for some guidelines
or best practices and can't find much info.

Thanks for the input!

.

Relevant Pages

  • Booting a poor-mans LAVC [was:Interesting cluster config "deadlock"]
    ... I managed an environment where a VAX with locally attached DSSI disks ... We also needed to retain cluster quorum. ... then tried to mount the shadow set with ...
    (comp.os.vms)
  • Re: Installing SQL205 in a cluster with mount points
    ... Most of the installation goes somewhere such as C:\Program ... Anyway, to summarise, the base drive isn't just for mount points as the ... am trying to set up a 2 node cluster with a single instance of sql2005. ... disks which are mounted in there. ...
    (microsoft.public.sqlserver.clustering)
  • Re: Adding a node to a one-node functional cluster
    ... Until then, both nodes will try and mount them, typically with bad ... Once the cluster service writes an intentionally mangled ... > on the disks, the OS thinks they are foreign partitions and won't try and ... > requires so the new node can see the LUNs as well as the old node. ...
    (microsoft.public.sqlserver.clustering)
  • Re: chicken and egg: shadowing and licenses
    ... > On my systems, I mount all disks, be it shadow sets or plain disks, ... The cluster wide logical name LMF$LICENSE points to this file. ... > And I get no license warnings during startup! ... I mount the cluster-wide disk from SYLOGICALS.COM and mount other disks ...
    (comp.os.vms)
  • Daily Report #4165
    ... The resultant cleaned cluster CMDs will ... well-understood host galaxy environment. ... The Nature of the Halos and Thick Disks of Spiral Galaxies ... ACS, NICMOS, and WFPC2 in parallel. ...
    (sci.astro.hubble)
Loading