Hardware advice


#1

Hi all,

I’m looking at purchasing a server to run some SciDB testing on, and, since the user guide says to consult the forum for advice, ended up here. I was looking specifically at Dell r720s - dual socket Xeon E5 boxes that can hold either 8 or 12 3.5" disks. Is that a reasonable server for SciDB purposes, or should I be looking at configurations with more or less disks per core? Roughly how much memory should be available per core? Finally, how important is HW RAID - would a typical deployment on a node like the above use JBOD + an instance per disk, or a smaller number of instances (perhaps just one) per node?

Thanks for any recommendations,
Serge


SciDB Hardware Recommendations
#2

Hello, Serge

Our recommendation is this:
1-2 CPU cores + 1 hard drive spindle + 2-4 GB ram = 1 scidb instance
That’s the recommended balance. Multiple instances can coexist on the same OS as processes - no need to worry about setting up virtual machines. We recommend you tell scidb to use replication and don’t put in any RAID. SciDB will handle the disk failure protection for you and instances will control disks independently.

Does this answer your question?


#3

Yes it does - thanks very much!


#4

Ah… about that: where in the manual is replication discussed?

I scanned the entire scidb-userguide-12.10 and found only one reference to a replicated array, and that was in the context of an afl operator intended for internal use
only.

George


#5

I made a mistake. I forgot that replication is a proprietary plugin feature provided by P4, not part of SciDB. My apologies for the confusion. For more information, please send email to info [at] paradigm4 [dot] com.


#6

What about SSDs vs HDDs?

We are working on an experimental cloud deployment in Amazon, and I was wondering how many SSDs we need, in the forumla:
1-2 CPU cores + 1 hard drive spindle + 2-4 GB ram = 1 scidb instance

Thanks,
Ohad.