ERP and in-memory workloads are some of the last bastions of “must have on-premise” workloads. Typically these workloads run on expensive hardware with huge amounts of memory. For those workloads, the relatively mundane specs of cloud instances have been a significant barrier. With the new Amazon AWS EC2 X1e memory-optimized instances, Amazon is offering hardware capable of servicing some of these workloads.
Amazon AWS EC2 X1e Instances
Here is the new AWS EC2 X1e instance lineup and their pricing:
|Model||vCPU||Memory (GiB)||SSD Storage (GB)||Dedicated EBS Bandwidth (Mbps)||$/hour|
|x1e.32xlarge||128||3,904||2 x 1,920||14,000||26.688|
|x1e.16xlarge||64||1,952||1 x 1,920||7,000||13.344|
|x1e.8xlarge||32||976||1 x 960||3,500||6.672|
|x1e.4xlarge||16||488||1 x 480||1,750||3.336|
|x1e.2xlarge||8||244||1 x 240||1,000||1.668|
|x1e.xlarge||4||122||1 x 120||500||0.834|
Powering these instances are Intel Xeon E7-8880 V3 (Haswell) generation CPUs at 2.3GHz. See the full specs on ARK. These CPUs allow for higher memory capacities albeit they are now a full two generations older than state of the art. The instances also have interesting options such as the x1e.xlarge with 4 vCPUs a staggering 122GiB of RAM. If you compare that to a c5.xlarge also with 4 vCPUs there is a staggering jump in memory. The c5.xlarge has 8GB of RAM while the r4.xlarge has only 30.5GB of RAM.
For networking, one can get up to 25gbps for the instance. There is also up to 14gbps of EBS bandwidth and up to 80K IOPS.
While 4TB may sound like a lot, but it is a mere fraction of where AWS EC2 is going. Amazon announced plans for machines with up to 16TB of RAM.
The bigger concern running this class of applications in the cloud is licensing. License costs can quickly dwarf the instance costs. Another key item is that one will spend about as much as a full system in one year with the AWS service even using reserved pricing (Over $133K/ year) at the 4TB level. ERP class applications must run 24×7 so the hardware plus operations costs are likely lower in a buy and operate model in under one year.