in my shop we very rarely create custom size devices, meaning we have standard symdev size (let's say 8.6G) and out of that we can create custom size metas. We try to tell our unix admins not to request one big 300G device but a couple of smaller metas which give you more i/o queues in the OS. Different operating systems have different LVM restrictions, some can only have so many devices of certain size in a volume group ..etc.
In our shop we have standardized on two device sizes: 8 GB and ~33GB. From these we build whatever metas are required, but try to limit the size to something reasonable. Then we build the Metas from whichever devices make the most sense.
We try to ensure that the Metas we provide are best matched to the sized required, but there are always occasions where a volume group contains multiple metas and/or devices. Usually we try to keep a single file system to a VG as long as there is any kind of performance concern. This helps us better manage where the data goes. That being said, most of the servers have a "slush" VG which has no special performance requirements and contains all the "rest" of the file systems.
This has worked well for us for a few years now despite having to educate and strongly influence the Unix Admins in the early days. We still have to remind them regularly of how to rephrase their request to better meet their own needs, but that isn't too painful any more :-)
dynamox
9 Legend
•
20.4K Posts
0
December 8th, 2009 06:00
Allen Ward
4 Operator
•
2.1K Posts
0
December 22nd, 2009 13:00
In our shop we have standardized on two device sizes: 8 GB and ~33GB. From these we build whatever metas are required, but try to limit the size to something reasonable. Then we build the Metas from whichever devices make the most sense.
We try to ensure that the Metas we provide are best matched to the sized required, but there are always occasions where a volume group contains multiple metas and/or devices. Usually we try to keep a single file system to a VG as long as there is any kind of performance concern. This helps us better manage where the data goes. That being said, most of the servers have a "slush" VG which has no special performance requirements and contains all the "rest" of the file systems.
This has worked well for us for a few years now despite having to educate and strongly influence the Unix Admins in the early days. We still have to remind them regularly of how to rephrase their request to better meet their own needs, but that isn't too painful any more :-)