JBOD server configuration and multiple storage locations

[awong-openx]

Hi,

Does anyone know if there is a limitation on the number of storage locations you can setup in Vertica? I have machines with 8 JBOD 300gb disks I would like to setup a data storage location for each disks. Is this a supported/recommended server configuration?

Thank you

[fsalvelt]

Hi,

I storage location is nothing more that a directory on disk ... So, I don't think there is a physical limitation in Vertica on how may you can have, however, I'd become an administrative nightmare if you create too many!

I'm not completely sure how JBOD works, but how are you going to separate temp storage from data storage if you only have one giant theoretical disk? Plus, if one of those drives fails, will you lose all of your data? That's scary.

Good luck!

[awong-openx]

Thank you replying

Jbod is simply a set of individual disks in a machine and won't be a one giant disk. My question is how does vertica handle disk failure from one of those data disk storage location as opposed to one big giant disk. In a multiple disk setup with the distributed projections, in theory, if one disk fail from a machine, its data should still be replicates on another node which should not cause loss of data.

[fsalvelt]

Hi,

If any of the disks on a single node fails then the whole node fails. Data is segmented or replicated across nodes assuming you have 3+ nodes and a ksafe level of at least 1. Each node needs to have identical hardware (to my understanding).

Are the 8 disks you mentioned in one node? Or are these on a separate computer (i.e. SAN)? If not, you'd probably have to have at least three nodes, each having 8 disks.

Hmm. Maybe I completely misunderstanding you

[awong-openx]

I will have 21 nodes with 8 x 300gb disks on each node. One disk will run the OS. No RAID at all just a bunch on individual disks.

[fsalvelt]

Hi,

You have 21 nodes? Nice!

I don't know about your hardware, but can the host remain up if one of the disks fail? If so, Vertica will keep the node alive as long the data it's storing is available. So I'd image if of the 8 disks on a host crashes and it has data, the node will go offline.

In hour test system we had a server lose a single drive but the node never went down. There was s storage location defined on it, but no data. It was a slower driver and Vertica will write to it last once the faster drivers fill up.

[awong-openx]

That is what I'm suspecting also that when a storage location disk fail the node will go down. That's what I'm trying to confirm. Anyone know?