OakBio Labs
Bare-metal Postgres at 200k QPS without breaking the audit.
OakBio's bioinformatics platform crunches genomic data for 90 academic and pharma customers. When their primary Postgres saturated a 96-core managed instance, they ran a 14-day evaluation against three hosting providers and chose Vintony's dedicated tier — partly for the IOPS, partly because the compliance pack was already in place.
Read p99
<3ms
Sustained at 200k QPS
Nightly job
1h 47m
Was 6+ hours under contention
Pages this quarter
0
From the database tier
HIPAA paperwork
9 days
BAA + audit handover
The problem
OakBio's reference dataset weighs in at 2.4 TB after compression. A single tenant's nightly variant-calling job touches 90% of it, parallelised over forty workers reading partitioned tables. On a managed Postgres, IOPS contention with co-tenants meant nightly jobs occasionally took six hours instead of two — and when they did, nineteen pagers fired across the customer base. The platform team needed deterministic IOPS, NVMe-class read latency, and they needed to keep their HIPAA-aligned audit-trail intact through the migration.
The migration
Vintony shipped two dedicated NVMe RAID-10 boxes — one Frankfurt, one London — within four hours of order. The OakBio platform team ran a 14-day shadow against their existing managed Postgres, replaying read traffic via pg-rewind and comparing p99 read-latency in the background. The shadow ran at 200,000 QPS for eleven days straight without breaking 3ms p99. Cutover was a Friday-evening logical replication catch-up window of 38 seconds, gated by their on-call SRE. The compliance side was unblocked by Vintony's pre-existing HIPAA BAA and the SOC 2 Type II report; their compliance officer signed off on the audit-trail handover the same week.
“We needed bare metal with NVMe RAID and a private VLAN, on a tight timeline. Vintony shipped it within four hours of order.”
What's next
OakBio is evaluating a third dedicated box in Singapore for an APAC pharma customer expected to onboard in Q4. Their platform team is also exploring Vintony's managed Postgres for the smaller per-tenant warm caches that don't need the dedicated tier.