Re: Hardware Requirements for RADOS Gateway Cluster

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Brice!

On 09/24/2012 05:58 PM, Brice Burgess wrote:
I am planning the system architecture around a ceph cluster [my first!].
While resources for setting up the basic cluster (MON+MDS+OSD) are
readily available*, I haven't come across anything outlining production
recommendations for a RADOS Gateway Cluster. Specifically;

1. Is it preferable to run the RADOS Gateway on a MDS machine [for
latency issues], or should the Gateway run on seperate hosts/VMs than
the ODS/MON/MDS servers?


It's entirely possible to run the gateway on an OSD/MON/MDS. I don't think we've done any kind of extensive analysis of where the best place to run one is. My guess is that it will only matter in very specific cases.

2. I've seen reference that multiple RADOS Gateway servers can be setup
as a cluster "to scale". Is this possible? Is there an
example/documentation for this? Is it just 2 standalone gateways with a
load balancer in front?


Yes! It's entirely possible to run multiple RGWs with a load balancer in front. I'm not sure how much documentation is out there for this yet, but it's something that customers of ours have successfully implemented.

My assumption is to provision a dedicated machine for the RADOS Gateway.
I'd treat this machine as a "front end proxy/caching server" meaning it
would have a lot of RAM for varnish/nginx and a low latency, high
throughput network connection to the ODS/MDS machines as well as one to
the public network.

Sounds good to me.  Other people might chime in with more suggestions.


Any thoughts and suggestions are welcome. Eventually I'd like to release
a whitepaper of this setup if we can hit the budget to actually
implement ;)

Some things you'll want to keep in mind:

You'll want to make sure that your testing (especially of small IO) is done across many buckets. Performance of a single bucket can be limited by the OSD where it's log is stored (or optionally you can turn the log off). You'll also want to keep in mind that RGW will likely incur larger performance overhead for small objects than large ones.

Thanks,
Mark
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Index of Archives]     [CEPH Users]     [Ceph Large]     [Information on CEPH]     [Linux BTRFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]
  Powered by Linux