All,
We have a development environment which supports quite a few projects. In order to support the typical development lifecycle we have a directory structure on our servers mounted from the filer which mean we consume 10 or so quota tree entries per project. These are all on different volumes to separate databases, archives/export areas and software etc. to enable us to manage space allocation/snapshots for each different type of use. The set-up has been working very well for two years or so.
What I perceive to be the issue with large disks, is for some of the volumes we will only need one or two disks to provide the required capacity (at 70Gb per disk for example = 140Gb with three disks approx just using simple arithmetic). This is below the optimum for performance on a raid group. The parity disk as a ratio is 1/3 of the volume which is inefficient by a long way. Given there is a limit of 255 quota trees per volume adding any more disks for performance/efficiency is a waste of space as the number of quota trees will be exceeded before the need for space is. As disks get larger the problem gets worse. I can see how this situation would not necessarily arise if you are an ISP but development is a different issue I feel.
The way round this may be to have what I have decided to call "virtual volumes". By this I would create a volume with sufficient disks in the raid group to give me both performance and efficiency. This would the be able to support a number of these "virtual volumes" across those disks. Each of those "virtual volumes" would also need to be adjustable in size either up or down and facilitate all other volume characteristics. This would then allow me to have an efficient system and also be able to manage the quota tree, placement and snapshot issues.
This would also allow me to make full use of say an F840. It's projected capacity I could never achieve in a clustered environment as the maximum number of shelves per head is still 16. The only way to reach the capacity the filer head can support is using larger disks.
Have I missed anything?
Is there a better way to resolve this other than the suggestion above?
Do you see it as a problem - or am I out on a limb here?
Thanks for any input Ray Griffith