High-speed non-volatile memory express (NVMe)-based flash array maker Pavilion Data appears to have hit a wall. The US-based startup had dreamt of being a leading player in the storage market with its high-performance NVMe-based HFA all-flash arrays. It was all set to brief IT journalists as part of the IT Press Tour in Silicon Valley this week, but pulled out on the eve of the event.
Pavilion succeeded in raising $45m in a funding round completed in January 2022. But it failed to raise any more and the leadership team seems to have thrown in the towel after failing to secure a buyer for the company at a price deemed acceptable. It was then that the investors pulled the plug.
Of its 100 employees, 96 will be laid off, with US staff getting two weeks’ severance pay and those in India one month.
Pavilion was among a flurry of startups that tried to ride the wave of the emergence of NVMe flash media around five years ago. They also included Apeiron, E8, Excelero, Mangstor and Vexata, all of which have now bitten the dust too.
What fundamentally pulled the rug from under Pavilion and other NVMe box suppliers was that the big storage array makers – Dell EMC, Hitachi, NetApp, IBM, HPE and Pure – got in on the game and made NVMe and its connectivity via network protocols mainstream.
Very fast with multi-protocol pretensions…
The design of Pavilion’s Hyperparallel Flash Array (HFA) was a 4U box with 72 NVMe solid-state drives (SSDs) driven by 20 parallel-mounted controllers. That’s five times more controllers than in a traditional storage array, and each had its own Intel processor.
That configuration allowed an HFA array to get throughput of 120GBps in storage area network (SAN) mode, 90GBps as network-attached storage (NAS) and 80GBps in object storage mode. In terms of input/output operations per second (IOPS), it could achieve 20 million IOPS in block storage SAN mode, 4.5 million IOPS as NAS. Latency was 100μs (microseconds) in block access and 170μs for file access.
Pavilion made great claims about the rapidity of access it could deliver. In a 2020 interview with LeMagIT, Pavilion’s partnership chief, Costa Hasapopoulos, said: “We offer all protocols on our machine. If you compare what is delivered per U we provide seven to 50 times more IOPS, seven to 10 times more throughput, and two to seven times more capacity in block mode than Dell EMC, NetApp and Pure Storage.
“In file access mode on reads, our access times are seven times more rapid and writes 12 times faster than those of Isilon and Vast Data, which claim to be the champions of NAS. In object mode, our reads are near to six times faster and writes nine times more rapid than those of Minio and OpenIO.”
…but not at the same time
LeMagIT noted that the HFA suffered from some shortcomings that didn’t get put right. One of those was that the array, although capable of block, file and object access, was not capable of working in these modes simultaneously and was eventually more a cluster of independent arrays in the same box.
In addition, the HFA was one of the first to be accessible (in SAN mode) from servers via NVMe-over-TCP and NVMe-over-ROCE. These protocols, which allow high-speed storage networking over Ethernet networks are much less costly than Infiniband, which was used at first with NVMe.
