Case study · Education · MigrationStudi kasus · Pendidikan · Migrasi
From Nutanix to Proxmox VE, one batch at a time.Dari Nutanix ke Proxmox VE, satu batch demi satu batch.
A state university in Indonesia decided not to renew its Nutanix licenses, as the platform no longer fit its future IT direction. Proxmox VE with Ceph was installed on the existing Nutanix hardware, and workloads moved across in a planned, three-phase migration.Sebuah universitas negeri di Indonesia memutuskan tidak memperpanjang lisensi Nutanix, karena platform tersebut tidak lagi sejalan dengan arah TI ke depan. Proxmox VE dengan Ceph dipasang di atas hardware Nutanix yang sudah ada, dan beban kerja dipindahkan lewat migrasi terencana dalam tiga fase.
Summary for managementRingkasan untuk manajemen
- Why: renewing Nutanix was not feasible for the university's future IT vision, so it chose an open-source platform instead.
- Goal: move to Proxmox VE on the hardware the university already owned, without losing any running service.
- Approach: with two Nutanix clusters, empty one first using zero-downtime live migration, reinstall its nodes with Proxmox VE and Ceph, then move workloads across in small batches.
- Result: 16 running VMs (4.2 TiB) migrated in 9 batches between September 3 and 12, 2025. Powered-off VMs were migrated and kept as restorable backups.
- Bonus: 25 overprovisioned VMs were rightsized, freeing CPU and memory on both platforms.
- Alasan: memperpanjang Nutanix tidak feasible bagi visi TI kampus ke depan, sehingga dipilih platform open-source.
- Tujuan: pindah ke Proxmox VE di atas hardware yang sudah dimiliki, tanpa kehilangan layanan yang berjalan.
- Pendekatan: dari dua cluster Nutanix, kosongkan satu lebih dulu dengan live migration tanpa downtime, pasang ulang node-nya dengan Proxmox VE dan Ceph, lalu pindahkan beban kerja dalam batch kecil.
- Hasil: 16 VM aktif (4,2 TiB) dimigrasikan dalam 9 batch antara 3 dan 12 September 2025. VM yang mati juga dimigrasikan dan disimpan sebagai backup yang bisa di-restore.
- Bonus: 25 VM overprovisioned di-rightsize, membebaskan CPU dan memori di kedua platform.
Three phasesTiga fase
The cluster to be repurposed had to be emptied before it could be rebuilt, so the migration ran as a relay between the two clusters.Cluster yang akan dialihfungsikan harus dikosongkan dulu sebelum dibangun ulang, sehingga migrasi berjalan seperti estafet antara dua cluster.
Simplified diagram. Cluster B, running on the same Nutanix hardware, is the one reinstalled with Proxmox VE; Cluster A is the source of the batch migration.Diagram disederhanakan. Cluster B, di hardware Nutanix yang sama, dipasang ulang dengan Proxmox VE; Cluster A menjadi sumber migrasi per batch.
How each phase was doneCara setiap fase dikerjakan
Phase 1: empty the cluster, with no downtimeFase 1: kosongkan cluster, tanpa downtime
- On-Demand Cross-Cluster Live Migration moved user VMs from Cluster B to Cluster A while they kept running.
- The feature needed newer software, so Prism Central and AOS were upgraded in three stages (AOS 5.20 to 7.0, Prism Central 2021.7 to 7.3) to stay compatible at every step, without downtime.
- Capacity was made room for by rightsizing 25 overprovisioned VMs to twice their one-week average usage, and exporting powered-off VMs as OVA backups.
- Prism Central itself was moved with its backup-and-restore feature.
- On-Demand Cross-Cluster Live Migration memindahkan VM pengguna dari Cluster B ke Cluster A tanpa mematikannya.
- Fitur ini butuh versi software yang lebih baru, sehingga Prism Central dan AOS di-upgrade dalam tiga tahap (AOS 5.20 ke 7.0, Prism Central 2021.7 ke 7.3) agar tetap kompatibel di setiap langkah, tanpa downtime.
- Ruang kapasitas disiapkan dengan me-rightsize 25 VM overprovisioned menjadi dua kali rata-rata pemakaian seminggu, serta mengekspor VM yang mati sebagai backup OVA.
- Prism Central sendiri dipindahkan dengan fitur backup-and-restore.
Phase 2: rebuild as Proxmox VE HCIFase 2: bangun ulang sebagai HCI Proxmox VE
- The three emptied Nutanix nodes were reimaged with Proxmox VE and joined into one cluster: no new servers were needed.
- Two network links per node in an active-backup bond: automatic failover without special switch configuration.
- Tiga node Nutanix yang sudah kosong di-reimage dengan Proxmox VE dan digabung menjadi satu cluster: tidak perlu server baru.
- Dua link jaringan per node dalam bond active-backup: failover otomatis tanpa konfigurasi khusus di switch.
Ceph storage designDesain storage Ceph
- Ceph on all three nodes, with monitors and managers on each, using every data disk (SSD and HDD) as OSDs.
- Separate CRUSH rules for SSD and HDD, so data lands on the intended media.
- Three pools: an SSD pool for performance VMs, an HDD pool for capacity VMs, and CephFS on HDD (with SSD metadata) for backups, ISOs and imports.
- LZ4 compression on the pools to stretch usable capacity.
- Ceph di ketiga node, dengan monitor dan manager di setiap node, memakai seluruh disk data (SSD dan HDD) sebagai OSD.
- CRUSH rule terpisah untuk SSD dan HDD, agar data ditempatkan di media yang dimaksud.
- Tiga pool: pool SSD untuk VM yang butuh performa, pool HDD untuk VM yang butuh kapasitas, dan CephFS di HDD (metadata di SSD) untuk backup, ISO, dan impor.
- Kompresi LZ4 di pool untuk memperbesar kapasitas efektif.
Phase 3: batch migration with automationFase 3: migrasi per batch dengan otomasi
- Running VMs were grouped into 9 batches, with smaller VMs first to prove the process.
- Each VM was exported as OVA and imported with a custom script that reads CPU, memory, disks and NICs from the OVF, applies VirtIO, assigns the next free VM ID and starts the VM.
- Two VMs that failed the OVA path were moved manually: disk converted to qcow2 on the source, then attached to a new VM on Proxmox VE.
- Powered-off VMs were migrated the same way, then backed up and protected so they can be restored on demand.
- VM aktif dikelompokkan menjadi 9 batch, dimulai dari VM yang lebih kecil untuk membuktikan prosesnya.
- Setiap VM diekspor sebagai OVA lalu diimpor dengan script khusus yang membaca CPU, memori, disk, dan NIC dari OVF, menerapkan VirtIO, memberi VM ID kosong berikutnya, dan menyalakan VM.
- Dua VM yang gagal lewat jalur OVA dipindahkan manual: disk dikonversi ke qcow2 di sumber, lalu dipasang ke VM baru di Proxmox VE.
- VM yang mati dimigrasikan dengan cara yang sama, lalu di-backup dan diproteksi agar bisa di-restore saat dibutuhkan.
Migration timelineLinimasa migrasi
| Date (2025)Tanggal (2025) | VMs migratedVM dimigrasikan | Disk sizeUkuran disk |
|---|---|---|
| Sep 33 Sep | 5 | 107 GiB |
| Sep 44 Sep | 4 | 334 GiB |
| Sep 66 Sep | 3 | 1,483 GiB |
| Sep 77 Sep | 2 | 1,477 GiB |
| Sep 1010 Sep | 1 (manual)(manual) | 391 GiB |
| Sep 1212 Sep | 1 (manual)(manual) | 496 GiB |
| TotalTotal | 16 | 4,287 GiB (4.2 TiB) |
Small VMs went first to prove the process; the largest VMs (up to 946 GiB) followed once it was stable.VM kecil dimigrasikan lebih dulu untuk membuktikan prosesnya; VM terbesar (hingga 946 GiB) menyusul setelah proses stabil.
ResultsHasil
Running workloads migrated to Proxmox VE in 9 batches over 10 days.Beban kerja aktif dimigrasikan ke Proxmox VE dalam 9 batch selama 10 hari.
when emptying the cluster in phase 1, using cross-cluster live migration.saat mengosongkan cluster di fase 1, dengan cross-cluster live migration.
rightsized to real usage, freeing resources on both platforms.di-rightsize sesuai pemakaian nyata, membebaskan resource di kedua platform.
Nutanix or VMware renewal coming up?Masa perpanjangan Nutanix atau VMware sudah dekat?
Your current hardware may be able to run Proxmox VE. A migration succeeds in the plan, not on the night: in a free 30-minute consultation we look at your clusters, capacity and the safest path.Hardware yang Anda miliki sekarang mungkin bisa menjalankan Proxmox VE. Migrasi berhasil di tahap perencanaan, bukan di malam eksekusi: dalam konsultasi gratis 30 menit, kita tinjau cluster, kapasitas, dan jalur paling aman.
Customer identity, hardware brands and identifying details are withheld for confidentiality.Identitas pelanggan, merek hardware, dan detail yang dapat mengidentifikasi tidak dicantumkan demi kerahasiaan.