kota at basho.com
Mon Aug 18 22:17:39 EDT 2014
All three plans would be good choice, depending on what you need. RAM
and disk size seems enough. It's just tradeoff - assuming your
software configuration is correct.
If you want high I/O throughput, faster disk and more nodes to
distribute I/O load. Or if you want cost efficiency per data capacity,
I'll recommend fewer nodes with high density. The size of each node's
disk might be related to the time of handoff, which is, time to get
back to healthy state after node/disk failure.
I would recommend to keep the number of buckets fewer, say at most hundreds .
On Mon, Aug 18, 2014 at 2:46 PM, Anton Khalikov <anton at khalikov.ru> wrote:
> Hello there
> As our Riak CS cluster grows, we went through several performance problems
> which caused me to think we went a wrong way with planning our hardware
> nodes configuration probably.
> If we want to have an s3 cluster with 30-40 Tb capacity, 50-100 millions of
> keys (files) stored on few thousands of buckets with n_val=3 which way to
> choose hardware configuration for nodes would be better:
> 1. more nodes with less memory and hdd capacity each. Let's say 18 nodes
> with 40 gb ram and 8 tb storage per node
> 2. less nodes with more memory and hdd capacity each. Let's say 12 nodes
> with 64 gb ram and 12 tb storage per node
> 3. even less nodes with even more memory and hdd capacity each. Well, for
> example 6 nodes with 128 gb ram and 24 tb storage each
> Or is there a dependency between total cluster memory and total cluster
> storage capacity? What is our maximum storage capacity if we have 768 gb ram
> in total in our cluster in other words?
> Best regards
> Anton Khalikov
> riak-users mailing list
> riak-users at lists.basho.com
Kota UENISHI / @kuenishi
Basho Japan KK
More information about the riak-users