Thanks Dmitri and Matt for the info, those are very helpful. The short version is great, will dive into the links more.
On Thu, Apr 25, 2013 at 2:16 PM, Dmitri Zagidulin <dzagidu...@basho.com>wrote: > Tom, > > In addition to Matt's links above, I would recommend to take a look at the > following pages: > > > http://docs.basho.com/riak/latest/references/appendices/Cluster-Capacity-Planning/ > and > http://docs.basho.com/riak/latest/tutorials/System-Planning/ > > The short version is: > * RAM is important, especially with Bitcask. I personally wouldn't give > nodes less than 16GB each, though I suppose they could run with 8. (The > reason I mention Bitcask, is because it stores all of its keys in memory, > and if you have too many keys and too little memory, you're going to run > out. See > http://docs.basho.com/riak/latest/references/appendices/Bitcask-Capacity-Planning/for > a calculator that can help you ensure how much RAM you need, if you > know how many millions of keys you're going to store, and how large each > one is) > > * Disk IO speed is very important, and is usually the bottleneck for > clusters. RAID or SSDs are helpful. If you're running on AWS, disk IO is > especially killer, so spring for sufficient Provisioned IOPS. > > * Network bandwidth is important (between the nodes). If your cluster has > the default replication setting of N=3, each read or write request to a > node results in potentially 3 network requests to other nodes (this is in > addition to background-level "gossip" between the nodes, as they try and > maintain consistency, perform read-repair and so on). So basically, use the > fastest network speed you can afford (that makes sense for your > application). If you can spring for Gigabit+, go for it. If the application > is especially time-sensitive and is under high load, it may make sense to > have 2 network adapters for each node -- one for external client requests, > and one for internal inter-node traffic. > > * If you're running on AWS, your minimum instance size should be m1.large, > though really, you should use xlarge. (And don't forget about those PIOPS). > > Does that help clarify? > > Dmitri > > > > > On Mon, Apr 22, 2013 at 7:11 PM, Matt Black <matt.bl...@jbadigital.com>wrote: > >> There's a lot of information in here: >> >> http://docs.basho.com/riak/latest/cookbooks/Linux-Performance-Tuning/ >> >> And the sister article, specific to AWS: >> >> http://docs.basho.com/riak/latest/cookbooks/Performance-Tuning-AWS/ >> >> As ever with Riak, the answer to your question depends on what you're >> doing :) >> >> >> On 23 April 2013 02:07, Tom Zeng <t...@intridea.com> wrote: >> >>> Hi, >>> >>> I am wondering if there are any docs/recommendations on server hardware >>> specs for production Riak instances - cores, clock speed, network >>> interfaces/bandwidth, hard drive raid vs ssd, and etc. >>> >>> Tom >>> Thanks, >>> -- >>> Tom Zeng >>> Director of Engineering >>> Intridea, Inc. | www.intridea.com >>> t...@intridea.com >>> (o) 888.968.4332 x519 >>> (c) 240-643-8728 >>> >>> _______________________________________________ >>> riak-users mailing list >>> riak-users@lists.basho.com >>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com >>> >>> >> >> _______________________________________________ >> riak-users mailing list >> riak-users@lists.basho.com >> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com >> >> > -- Tom Zeng Director of Engineering Intridea, Inc. | www.intridea.com t...@intridea.com (o) 888.968.4332 x519 (c) 240-643-8728
_______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
