I am used to traditional SANs. On a traditional SAN, we're able to create tiers of storage (fast disks, slow disks), and then allocate volumes from those tiers of storage. This way a device that needs ultra low latency and ultra high throughput can get volumes built out of flash disks, and a data warehouse can get volumes built on slow spinning media, etc. Now obviously this is for block storage, but I am trying to figure out if gluster supports anything similar.
I've been reading through the Gluster docs, and I can't seem to find anything that directly supports classes of service.
If I deploy 9 storage nodes with 10 disks each, and the first 3 have "fast" NVME disks, the next 3 have "medium" enterprise SAS disks (15k rpm), and the last 3 have "slow" 7200rpm SATA disks, would gluster support providing different storage pools for each "class" or "tier" of service? I know a single pool can be hetereogeneous, but I am looking for separate pools.
The only thing I have found so far are mentions of "tiering" inside the gluster docs, but in their parlance they define "tiering" as a single pool with multiple types of disks in it, where slow disks act like cold storage for data that is rarely accessed, and fast disks act like caches.
Does gluster support classifying tiers of storage and making those tiers available to different endpoints?
Answer
Yes, you can create these kind of different 'classes'. What you would do with Gluster, is create a volume (basically network filesystem, share or export) that has all the bricks on one type of media, and an other volume with bricks on an other type.
Many users name the volumes after projects or departments, but there is no reason that you could not call a volume 'fast', 'medium' or 'slow'.
The tiering feature in Gluster is not what you are looking for. In addition it is not stable enough for most users, and slowly on its way out.
No comments:
Post a Comment