We are planning to install TD on our VM estate replicating our Development system. Although our development is on 14.10 we are planning to install 15.10 on VM and test our application on it. It is not going to be dedicated drives but looking at SAN option.. Amount of data is going to be about 8 TB...
1. Has anyone first replaced your dev / test environment with TD running on a complete VM infra?
2. How has been the performance specifically with SAN or dedicated drives?
3. What were your criteria for measuring minimum performance although it might be dev/test
4. Assuming if we are up and running the Teradata instance on a VM, Can i use the DATAMOVER to move the data between physical 14.10 to VM 15.10.. can i move data to a diff config system ie from 1 node physical to 2 node system ( 2 Virtual node system)
5. Else can i use a traditional arcmain backup recovery solution (assuming my arcmain client is 15.10)
Anything that you would want to share over and above.
1 - YES , In fact we have some customers running production level work on our TDVM solution
2 - Our reference standard is to use either dedicated drives or a dedicated array specifically for Teradata DATA storage in order to ensure performance across all nodes especially for an MPP deployment. A SAN is frequently used in a VMware/vSphere environment in order to share resources amongst many ESXi hosts and VMs. Keep in mind that cloud environments were designed to share resources and even over provision and allow vSphere to load balance as needed. Load balancing a TDVM instance is not recommended as this will definitely impact performance. We have seen dedicated SANs perform just fine however as soon as other VMs or systems are deployed using the same SAN performance is hindered.
3 - A difficult question to answer since all deployments and work loads are different. I always recommend a preSales engagement to analyze the goal and even recommend a POC first when possible.
4 - Yes you certainly can do that.
5 - A full archive and restore is also supported.
We have many customers that are doing exactly what you are describing. A single node SMP is the easiest and quickest to deploy but for 8 TB I would recommend at least an 3 vNode MPP system. Using a Dell R730xd(our reference standard) with internal drives in RAID 1 pairs this could easily be a fully functional Dev/Test environment. We use a standard VM size for our TPA nodes and normally deploy up to 4 VM TPAs on a single R730xd but servers of similar configuration are being used by customers with good results. The bottlenecks that we find are either in the bynet network interconnect (running larger MPP systems and connecting multiple R730xd nodes or similar together) or in the storage subsystems and a SAN unless dedicated is a frequent bottleneck. A Teradata system is the Ferrari of Databases and as such demands the most from the network interconnect and the storage subsystems. Take a look at our purpose built platforms and you'll see network interconnect cabling that sums up to 40/60GB and the storage is usually a dedicated diskarray attached to two nodes. If you want to replicate the performance then try to replicate the hardware as best as possible which is what we did. That said our network interconnect for TDVM is 2 separate dedicated 10GB networks for bynet and dedicated RAID 1 pairs. RAID 5/6 are a considerable performance impact although we do have customers using them and are satisified as they recognize the impact is there and accept the difference. This is a software only solution and you own the hardware so it is up to you to provide the level of performance that you want. The nice part is you can deploy and redeploy frequently, test, until you find the configuration that works for you.
Mark Ennis - TDVME Architect
Thank you Mark for your response.
You made a statement:
"We use a standard VM size for our TPA nodes and normally deploy up to 4 VM TPAs on a single R730xd "
What is your definition of standard size?
When you say 4 VM TPA's is it 4 individual SMP nodes ( treated as 4 different TD systems) or is my understanding wrong?
for 8 TB if you say i build a single VM ( 3 vnodes ) acting as an MPP system , how would we determine the number of amps needs or is it preconfigured? Assume if i determine 12 Amps/vnode ( in total 36 amps ) then how many #cores should i opt to liscense for? example 36 cores should i consider( 1 core / amp ) or the cores doesnt matter for the # of amps..
"our network interconnect for TDVM is 2 separate dedicated 10GB networks for bynet and dedicated RAID 1 pairs"
For a Single VM MPP ( 3 vnodes ) what whould be the Bynet bandwidth?
Sorry if my questions are silly but iam trying to adapt to the VM world :)