Senior Network Engineer (L3) at rackbank, Raipur

Open. Posted .

GPU cluster network fabric design and operations span BGP underlay, EVPN VXLAN, RoCEv2 and InfiniBand, with automation, telemetry and production incident ownership. The role is based in Raipur and includes planned night windows and on-call rotation.

Apply on rackbank's site for Senior Network Engineer (L3)

Applying opens the employer's page. Want jobs like this sent to you each morning? Open Elip.

Key facts
LocationRaipur, Chhattisgarh
Work modeOn-site
Experience4–8 years
PayNot stated by the employer
RoleNetwork Engineer
Skills asked forBGP, TCP/IP, Linux networking, Python, Change management
Posted30 Oct 2026

Elip's summary

Builds high-performance network fabrics for a data center company delivering cloud and colocation services. Designs and tunes GPU cluster networks using BGP, EVPN VXLAN, RoCEv2 and InfiniBand, and automates fabric provisioning and validation. Applies Python and network telemetry to troubleshoot congestion and production incidents.

The job, in the employer's words

Company Overview:

RackBank Datacenters is a fast-growing and trusted data center company committed to powering digital transformation with secure, scalable, and efficient IT infrastructure solutions. With state-of-the-art data centers in India and a strong client base across multiple industries, RackBank is focused on delivering next-gen cloud and colocation services.

Our Vision:

To build India’s most reliable and trusted data center and digital infrastructure ecosystem, empowering enterprises and governments with secure, scalable, and future-ready infrastructure.

Role Overview:

This is a senior individual contributor role that owns the high performance network fabric connecting our GPU clusters at Raipur. You will design it, build it, tune it, automate it, and own it when it breaks.

A GPU cluster is only as fast as the fabric underneath it. The difference between a cluster that delivers what was committed and one that quietly runs at sixty percent of it is almost never the GPUs. It is congestion control, buffer and queue behaviour, a priority flow control or ECN setting chosen without evidence, an MTU mismatch, or a topology that did not anticipate the traffic pattern of a large collective operation. Finding that class of problem, fixing it, and then building the tooling that stops it recurring is the core of this job.

What you will own

You are responsible for the fabric at Raipur from the moment a design is drawn to the moment a customer workload is running on it at the throughput we committed to. In practice that covers:

  • The BGP underlay and the EVPN VXLAN overlay across the leaf and spine fabric.
  • The lossless Ethernet and InfiniBand fabrics that carry GPU traffic, including their tuning and benchmarking.
  • The storage, management and out of band networks that sit alongside them.
  • The automation that provisions and validates all of it, and the telemetry that tells us something is wrong before a customer does.
  • The acceptance testing that signs a pod off, and the documentation and runbooks that let somebody else operate what you built.

Key Responsibilities:

Fabric design and build out

  • Design and bring up leaf and spine Clos fabrics for new GPU pods, including BGP underlay, EVPN VXLAN overlay, and the oversubscription, failure domain and scaling decisions that sit behind them.
  • Configure and validate RoCEv2 lossless Ethernet fabrics, covering priority flow control, explicit congestion notification, traffic classes and buffer allocation.
  • Bring up and operate InfiniBand fabrics where the cluster design calls for them, including subnet manager configuration and UFM.
  • Specify and verify optics, cabling plans and port maps with the data centre operations team, and resolve physical layer faults before they turn into performance mysteries.
  • Design and run the acceptance test plan that signs a pod off, so a cluster is proven before a customer is given access to it rather than afterwards.
  • Keep the storage, management and out of band networks designed and documented to the same standard as the GPU fabric.
  • Contribute to design reviews for new phases of the campus, and feed real operational experience back into the next design rather than repeating it.

Performance and congestion tuning

  • Tune priority flow control, explicit congestion notification, DCQCN, quality of service policy and buffer configuration for AI and HPC traffic patterns.
  • Benchmark cluster communication performance using NCCL and collective operation tests, and investigate the difference between expected and achieved results.
  • Maintain MTU, flow control and configuration consistency across every device in a fabric, and detect drift automatically rather than by inspection.
  • Analyse switch counters, queue depth, pause frames and discard statistics to find congestion before a customer reports it.
  • Work with the cloud platform team on how scheduler placement, topology awareness and job sizing interact with fabric behaviour.
  • Record what you learn from each tuning exercise so the next pod starts from a better baseline than the last one did.

Automation and telemetry

  • Replace manual switch configuration with templated, version controlled provisioning, so that two engineers building the same pod produce the same result.
  • Build automated validation and pre handover test suites that verify a fabric against its intended design rather than against somebody's memory of it.
  • Stream telemetry into dashboards and alerting that the operations team genuinely uses, and tune alerts until they are trusted rather than ignored.
  • Maintain a reliable source of truth for topology, addressing and inventory, and keep documentation current as a condition of closing work.
  • Write runbooks for the failure modes you have seen, so the next engineer does not start from zero at two in the morning.

Production operations

  • Respond to fabric incidents, lead root cause analysis, and close the loop with a fix or a documented control rather than a restart.
  • Plan and execute changes on a live fabric under change management, including rollback plans that have genuinely been thought through.
  • Review capacity, link utilisation and error trends, and raise capacity needs before they become constraints.
  • Plan and carry out firmware and network operating system upgrades with minimal customer impact.
  • Participate in a shared on call rotation and in planned night windows during cluster turn up.
  • Manage escalations with switching and NVIDIA ecosystem vendors, supported by your own evidence rather than by their assurances.

Customer and team support

  • Explain fabric behaviour, design choices and performance findings to customers' machine learning and platform teams, both in writing and on calls.
  • Support pre sales and solution teams on cluster sizing, topology and feasibility questions.
  • Mentor junior engineers, review their work, and raise the standard of documentation and process across the team.
  • Help interview and assess the engineers who join after you.

Key Requirements:

  • 4-8 years of hands on experience in data centre, cloud or service provider networking, in production environments rather than only in labs.
  • BGP configured, operated and debugged in production. You should be able to talk about route policy, path selection and convergence behaviour, and about a specific day when it did not behave as expected.
  • Strong layer 2 and layer 3 fundamentals and solid TCP/IP knowledge. You can take a packet capture and reason from it rather than only escalate it.
  • Practical Linux networking: interfaces, routing, namespaces and the standard diagnostic toolkit, on Red Hat or Ubuntu based systems.
  • Scripting ability, with Python preferred, applied to real configuration, validation or troubleshooting work rather than held as a certificate.
  • Experience working to change management and incident process in an environment where mistakes have customer impact.
  • Clear written technical communication. A large part of this job is explaining fabric behaviour in writing to people who are not network engineers.
  • Willingness to be based in Raipur, to work planned night windows during cluster turn up, and to share an on call rotation.

Strongly preferred:

  • EVPN VXLAN leaf and spine fabrics designed, deployed or operated in production.
  • RoCEv2 and RDMA, lossless Ethernet, priority flow control, explicit congestion notification, DCQCN, and quality of service and buffer tuning.
  • InfiniBand exposure, including HDR or NDR generations, subnet manager and UFM, or equivalent HPC fabric experience.
  • Hands on time with NVIDIA Spectrum or Cumulus Linux, Arista EOS, Cisco NX OS, Juniper Junos, or SONiC.
  • Network automation applied in production: Ansible, Nornir, NetBox, Jinja templating, gNMI, or a streaming telemetry stack.
  • Experience in a data centre, colocation, GPU cloud, hyperscaler or HPC facility environment.
  • Certification that demonstrates self direction: CCNP, CCIE written, JNCIP, NVIDIA networking certifications, or RHCE.
  • Familiarity with optical transport, structured cabling standards and high density rack environments.

Education:

A degree in electronics, communications, computer science or a related engineering discipline is the common route into this work and it is welcome. A diploma or ITI qualification combined with relevant certification and genuine production experience is equally acceptable to us, and we will assess both the same way.

About rackbank

Headquartered in Indore, founded in 2013.

People in Elip's index who work here
73
Typical time at rackbank
20 months
Typical career length so far
3.3 years

Largest teams

  • Information Technology12 people
  • Engineering10 people

Totals from public professional profiles in Elip's index; any group under ten people is left out.

All open jobs at rackbank

Similar open jobs

all jobs on Elip

About this page

Elip lists jobs that employers publish on their own careers sites, never copies from job boards. This one comes from rackbank's listing on Keka, posted 30 Oct 2026. The summary is Elip's; the description is the employer's. When the job closes, this page says so.