[MarkLogic Dev General] marklogic and amazon web services
mike at blakeley.com
Fri Jan 20 09:21:19 PST 2012
I've done some testing and evaluation work on AWS. CPU performance is acceptable if you get a newer Xeon-based VM, but less so if you get one of the older Opteron VMs.
Disk-heavy workloads can easily become I/O-limited, even with large numbers of EBS volumes in a RAID-0. That doesn't matter much for small databases, but when you are targeting large-scale databases it becomes a problem. It helps to limit your forest size so that the largest merges will still complete in about 3-hr, but that reduces the capacity per host and drives up costs.
Local disks (ephemeral storage) offer more predictable performance, but present their own problems around reliability. One attractive model is local disk + forest replication + database replication to another data center.
On 20 Jan 2012, at 06:55 , Jakob Fix wrote:
> I just wanted to get a couple of opinions of using AWS and MarkLogic
> for production work. Are people happy with the performance and more
> particularly with the I/O? There was quite some excitement last year
> about deploying ML via AWS. What's the experience one year later?
> General mailing list
> General at developer.marklogic.com
More information about the General