In my last blog post I’ve mentioned that in a joint EMC/VCE project we’ve successfully validated virtualized HANA on EMC ScaleIO. For those of you interested in getting a better understanding on how to implement HANA on VMware vSphere and how we configured the environment to run HANA on top of EMC ScaleIO: here we go…
We’ve done the testing in the same environment we’ve used for the physical implementation and that was described in detail in the white paper “Storage Configuration Best Practices for SAP HANA Tailored Data Center Integration and EMC ScaleIO Converged Infrastructure”. For each HANA node we configured one LUN per Data or Log device and ran it under VMFS. To achieve the HANA KPIs, the Log devices need to be built with SSDs or flash drives.
As you can see in the figure below, the primary and secondary MDMs were configured to run on the two Cisco C240 servers, the tie-breaker was on one of the C460s. The ScaleIO Data Clients (SDC) reside in the hypervisor exposing the LUN as a local device. All ESXi hosts offered a ScaleIO Data Server (SDS) in a dedicated ScaleIO VM in order to access storage volumes via VMFS. This means we had a fully converged configuration with both, SDC and SDS on each ESXi host. Btw, I’ve also seen customers who prefer running ScaleIO in a two-layer configuration, where a group of servers is running SDCs and another distinct group is running SDSs. In a way, this deployment is similar to a traditional external storage system. Applications run in one layer, while storage is in another layer. For more please take a look at my blog post here.
Now, let’s take a look at the detailed lab view. We’ve used the HANA VMs (vHANA1 – vHANA8) to run in different use cases. In a Multi-VM setup we showed that you can run 6 HANA systems on 2 ESXi servers and still fulfill all relevant SAP HANA TDI KPIs for productive use. Furthermore we took a look at 6+0 scale-out, VMware HA, vMotion and DRS.
We created 2 ScaleIO storage pools. One pool for the SAS drives and one for the SSDs (or alternatively with flash drives). For each ESXi server we mirrored 2 disks for o/s boot.
The storage configuration diagram shows the o/s, Log and Data vmdks for vHANA1 – vHANA8. You can also see the used drive slots.
During our validation we’ve used the following network configuration:
Here are some network related tweaks we configured to maximize performance across the networks, particularly the internal ScaleIO, and HANA internode networks:
- Nexus 3064-T Switch – To facilitate redundancy and aggregate network ports:
- Port-Channels were built within each switch
- Virtual Port Channels (VPC’s) built inter-switch
- Within vSphere :
- Large Receive Offload (LRO) were turned off
- Link Aggregation Control Protocol (LACP) trunks built to aggregate network interfaces
- Load Balancing enabled within the LACP trunk
- Within SUSE Linux set the length of the transmit queue for the network interface. A high value is recommended for server connected over the high-speed connection that perform large data transfers.
- txqueuelen 10000, e.g. ifconfig eth1 txqueuelen 10000
As already mentioned we also successfully validated VMware HA, vMotion and DRS.
For those of you who want to see how our validation looked like in the vShpere Web Client, the vSphere ScaleIO Plugin or the SAP HANA Administration Console, I’m adding the following 3 screenshots:
Want to proof? As announced at EMC World, ScaleIO is now available freely for non-productive use. With all features, for an unlimited time and unlimited capacity. For the download navigate to http://www.emc.com/scaleIO.