Case study · Banking · VirtualizationStudi kasus · Perbankan · Virtualisasi
Old hardware, new private cloud.Hardware lama, private cloud baru.
A regional bank in Indonesia refreshed its production cluster with a new solution, leaving more than 10-year-old hardware idle. This project gave that hardware a second life as a 13-node Proxmox VE cluster for development.Sebuah bank daerah di Indonesia meremajakan cluster production-nya dengan solusi baru, sehingga hardware berusia lebih dari 10 tahun menganggur. Proyek ini memberi hardware tersebut umur kedua sebagai cluster Proxmox VE 13 node untuk development.
Summary for managementRingkasan untuk manajemen
- Before: the hardware had run production on VMware for over 10 years. After the bank refreshed production with a new solution, it sat unused, and parts of it were faulty.
- Now: the same hardware runs a 13-node Proxmox VE cluster for development, with shared SAN storage and high availability for VMs.
- Value: the existing hardware investment gets a second life, on an open-source platform without subscription costs.
- Handover: standard VM templates and hands-on training for the bank's IT team.
- Sebelum: hardware ini menjalankan production dengan VMware selama lebih dari 10 tahun. Setelah bank meremajakan production dengan solusi baru, hardware ini tidak terpakai, dan sebagian komponennya bermasalah.
- Sekarang: hardware yang sama menjalankan cluster Proxmox VE 13 node untuk development, dengan shared storage SAN dan high availability untuk VM.
- Nilai: investasi hardware yang sudah ada mendapat umur kedua, di atas platform open-source tanpa biaya subscription.
- Serah terima: template VM standar dan pelatihan langsung untuk tim TI bank.
The challengeTantangan
After more than 10 years running production on VMware, the blades and SAN arrays were replaced by a new production solution. The bank wanted to keep using them rather than leave them idle.Setelah lebih dari 10 tahun menjalankan production dengan VMware, blade dan storage SAN digantikan solusi production yang baru. Bank ingin tetap memanfaatkannya, bukan membiarkannya menganggur.
The initial health check found memory errors on three nodes, one node failing to initialize, a failed power supply on a storage array, and a degraded disk.Pemeriksaan awal menemukan error memori di tiga node, satu node gagal inisialisasi, power supply storage yang gagal, dan satu disk yang degraded.
Step one: make the hardware trustworthyLangkah pertama: membuat hardware layak dipercaya
A cluster is only as reliable as its weakest node, so every fault was resolved before installation, using parts already on site.Cluster hanya sekuat node terlemahnya, sehingga setiap kerusakan diselesaikan sebelum instalasi, dengan komponen yang sudah ada di lokasi.
FoundDitemukan
- Memory errors (failed scrub, ECC limit reached) on three nodes
- One node failing to initialize
- Failed power supply on a storage array
- One degraded disk in the second array
- Error memori (scrub gagal, batas ECC tercapai) di tiga node
- Satu node gagal inisialisasi
- Power supply storage gagal
- Satu disk degraded di storage kedua
ResolvedDiatasi
- Faulty memory modules removed
- Node moved to another chassis slot
- Power supply replaced with one from an inactive identical unit
- Disk swapped with a healthy one from another inactive unit
- Modul memori bermasalah dilepas
- Node dipindah ke slot chassis lain
- Power supply diganti dari unit sejenis yang tidak aktif
- Disk ditukar dengan disk sehat dari unit lain yang tidak aktif
ArchitectureArsitektur
Every node reaches both storage arrays through two independent Fibre Channel fabrics. If a port, cable, switch or controller fails, storage stays reachable through the other path.Setiap node menjangkau kedua storage lewat dua fabric Fibre Channel yang independen. Jika port, kabel, switch, atau controller gagal, storage tetap terjangkau lewat jalur lainnya.
Simplified diagram. Fabric A connects port 1 of every node to controller A of both arrays; Fabric B connects port 2 to controller B.Diagram disederhanakan. Fabric A menghubungkan port 1 setiap node ke controller A kedua storage; Fabric B menghubungkan port 2 ke controller B.
The solution, layer by layerSolusi, lapis demi lapis
Storage arraysStorage
- Each array configured with RAID 5 and one hot spare, following the vendor's recommendation and using all available disks.
- Thick-provisioned LUNs as a safeguard against overprovisioning and running out of space.
- LUN masking so only the cluster's hosts can see the LUNs.
- The second array was cleaned up and reconfigured rather than factory-reset, to avoid the risk of losing its license.
- Setiap storage dikonfigurasi RAID 5 dengan satu hot spare, mengikuti rekomendasi vendor dan memakai seluruh disk yang ada.
- LUN thick sebagai pengaman dari overprovisioning dan kehabisan kapasitas.
- LUN masking agar hanya host cluster yang dapat melihat LUN.
- Storage kedua dibersihkan dan dikonfigurasi ulang, bukan di-reset pabrik, untuk menghindari risiko lisensi terhapus.
SAN fabricFabric SAN
- Two separate FC fabrics: port 1 of every node on Fabric A, port 2 on Fabric B.
- Single Initiator Multiple Target (SIMT) zoning: one zone per host port, so a misbehaving host cannot affect the others.
- Consistent naming for aliases and zones, making troubleshooting straightforward.
- Dua fabric FC terpisah: port 1 setiap node di Fabric A, port 2 di Fabric B.
- Zoning Single Initiator Multiple Target (SIMT): satu zone per port host, sehingga host yang bermasalah tidak mengganggu host lain.
- Penamaan alias dan zone yang konsisten agar troubleshooting mudah.
Proxmox VE hostsHost Proxmox VE
- Proxmox VE installed on each node's local RAID 1 boot volume.
- Linux multipath combines the paths to each LUN into one device, loaded at boot so disks are identified consistently across reboots.
- Shared LVM (thick) on top of the multipath devices: both arrays appear as cluster-wide storage on all 13 nodes, with snapshots enabled through volume chains.
- Proxmox VE dipasang di boot volume RAID 1 lokal setiap node.
- Linux multipath menggabungkan jalur ke setiap LUN menjadi satu perangkat, dimuat sejak boot agar disk teridentifikasi konsisten setiap reboot.
- Shared LVM (thick) di atas perangkat multipath: kedua storage tampil sebagai storage cluster-wide di 13 node, dengan snapshot melalui volume chain.
Network and clusterJaringan dan cluster
- Two Ethernet links per node in an active-backup bond, with link monitoring and delays that avoid needless failover on brief link flaps.
- The cluster link runs over the bond, so one failed path does not break quorum.
- Dua link Ethernet per node dalam bond active-backup, dengan pemantauan link dan jeda yang mencegah failover yang tidak perlu saat link berkedip sesaat.
- Link cluster berjalan di atas bond, sehingga satu jalur gagal tidak memutus quorum.
Operations and handoverOperasional dan serah terima
- High availability for VMs: protected VMs restart on another node if their node goes down.
- A Linux VM template with cloud-init and guest agent, used through full clones.
- Hands-on training for the bank's IT team after installation.
- High availability untuk VM: VM yang dilindungi berjalan ulang di node lain jika node-nya mati.
- Template VM Linux dengan cloud-init dan guest agent, dipakai melalui full clone.
- Pelatihan langsung untuk tim TI bank setelah instalasi.
ResultsHasil
Existing compute and storage hardware repurposed for development.Hardware compute dan storage yang ada dimanfaatkan ulang untuk development.
Retired production hardware now serves the development team on a current, open-source platform.Hardware production yang sudah pensiun kini melayani tim development di platform open-source yang mutakhir.
Dual fabrics, multipath and HA protect against single failures.Dual fabric, multipath, dan HA melindungi dari kegagalan tunggal.
Retiring old servers and SAN?Hardware lama akan dipensiunkan?
Hardware replaced in production may still have useful years ahead for development, testing or disaster recovery. In a free 30-minute consultation we review what you have and how to give it a second life.Hardware yang diganti di production mungkin masih punya umur berguna untuk development, testing, atau disaster recovery. Dalam konsultasi gratis 30 menit, kita tinjau apa yang Anda miliki dan cara memberinya umur kedua.
Customer identity, hardware brands and identifying details are withheld for confidentiality.Identitas pelanggan, merek hardware, dan detail yang dapat mengidentifikasi tidak dicantumkan demi kerahasiaan.