L32sv2 vs DS5v2 on Azure

For deploying huge Elasticsearch which Azure VM SKU is better?

DS5v2 – Intel Xeon processor
RAM: 56 GB
vCPU(s): 16
Managed Disk: Premium SSD

L32s v2: AMD EPYC 7551 processor
RAM: 256 GB
vCPU(s): 32
Local NVMe premium SSD disks attached to VMs (ephemeral)

By default, I was assuming we’d have to use managed disks as we couldn’t rely on the local temporary storage. That could get blown up at any time. Attached managed storage seems to be the way most people talk about supporting this scenario online.

The Lsv2-series has both temporary storage and NVMe disk. The spec sheet for it talks about it being ideal for “Big Data, SQL, and NoSQL databases.”. Which seems to fit our problem space.

We wanted to know if anyone tried Lsv2-series?

We run Elastic Cloud Hot HighIO nodes on Azure on lsv2 see here

For a single node you would probably use the following.

Standard_L8s_v2 8 vCPU 64 GB RAM

If you were going to virtualize you could use larger and then containerize.

Thanks @stephenb for the reply.

  • Can you provide more detail here why Lsv2, since Local NVMe disks are ephemeral?
  • How do you handled reboots or planned maintenance
  • Hardware failures
  • widespread outage

Hi @porscheme

Perhaps you should learn a bit more about Elasticsearch which is a distributed, resilient data store in its own, with replicated data etc.. and automated snapshots, it is highly resilient by design.

Elastic Cloud runs 20K+ cluster for some of the largest customers in the world.

This is a hosted platform most of those details are "Under the covers" ... we have full automation so all those things, HW failures, maintenance etc. Our SRE team takes care of all that. and we constantly are evaluating the best HW and Configurations to optimize cost and performance for our cloud customers.

This is the whole point of Elastic Cloud ... we do take care

Take a look at this it is the shared responsibility matrix.

Also looks like you are Azure Based and we just released and even deeper integration there

If you are concerned about Major Cloud provided outages like a whole region going down, you can chose to run in multiple regions with Cross Cluster Replication.

All that said... if you want you can run it yourself you certainly can!

Details are missing...all sales pitch.

Anyways, what is your compliance story?

Perhaps someone else is better suited to help you like @warkolm

  • Can you provide more detail here why Lsv2, since Local NVMe disks are ephemeral?

Elasticsearch is designed to be highly distributed high speed data store at scale. Local disk with primaries and replicas stored on different nodes (across AZs) + snapshot provide HA and the data, node and cluster level while providing the best performance. That is why we recommend local disk / SSD. You can certainly run on attached / managed storage but in general you will not get the same performance, but you certainly can.

Elastic Cloud runs 1000s of high performant resilient clusters, all run on local SSD / Disk (with the exception of the new Frozen data tied that is backed by S3, cache is local SSD)

Security and compliance.

You have not provided information about the use case ,the expected data volumes or the amount of traffic the cluster would need to support and without that it is hard to recommend hardware. It would help if you could provide some additional details and context.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.