First, let me say that switching to linstor with drbd-reactor was one of the best decisions we ever made. On our MySQL database servers, we run about 100 separate DRBD disks, each with its own ext4 filesystem and its own instance of MySQL. Each MySQL instance is for a different hosted customer. This gives us huge flexibility with maintaining instances for separate customers, It also has performance and availability advantages over using a single DRBD disk.
Now we are getting ready to deploy an FTP cluster (2 diskfull nodes and a tiebreaker). We’ve had such great success with our database servers that we are considering having a separate DRBD disk and filesystem for each of our 400-ish customers. Is that a good idea or a bad one? (Please say good.)
Hi Forbin! Thanks for the glowing endorsement firstly! Realistically, we haven’t tested something with that many resources for an active workload and it would depend on how many of those resources are active.
DRBD will support that many resources; It’s just about if there’s a capable enough I/O subsystem and Network to be able to handle that amount of traffic. Theoretically it would work, so I’d say it’s a good idea. Here’s a link with details about DRBD’s limits in terms of that, which this falls well within:
Please let us know how it goes for you, and feel free to reach out here for any other questions/concerns.
Thanks, Arty, I’ll peruse the article. Based on our experience with the workload, the hardware will not have any trouble keeping up. It worked fine on the old server with 7200 rpm spinners. The new server has more computing power, and storage is NVME. My only concerns are about whatever limitations DRBD itself may impose.
From what I understand, we (LINBIT) have had a user with ~950 production resources active at once. We have tested 2400 internally, with success, but no load other than creating them and bringing them up/down.
You might want to check out drbdmon a command line utility included with the drbd-utils package for managing many resources.