Panasas seeks partners as it faces up to public cloud challenge

Profile. A briefing meeting with HPC storage supplier Panasas’ chief exec in April revealed a company intent on broadening its partner base, building a lower entry-level system and working on understanding how it could partner with public cloud suppliers.

Panasas’ CEO is Tom Shea and he got the top job in September last year, after he was promoted from the COO position. 

Tom Shea.

Since then, Panasas has joined Securities Technology Analysis Center (STAC), the financial benchmarking organisation, and Shea has hired some new execs. Brian Petersen took on the COO role in November, 2020. Brian Reed became VP for Products and Alliances in February. Todd Ruff was hired as VP for marketing in March as was Richio Aikawa who joined as director of Vertical Solutions. Aikawa has a focus on the financial market and looks after the STAC area.

Shea says the prior move of the PanFS software to a Linux base “opened up new partnership possibilities.” 

The new exec hires are part of the reason that Panasas is apparently “a much more partner-centric company than before.” However, although he says it is his top priority, Shea had no new partnership news to share when we spoke.

The CEO says Panasas supplies its storage into the AI and machine learning markets. Knowing that GPU servers are a prominent feature of that market, B&F asked Shea about partnering with Nvidia. He wouldn’t be drawn, but did say: “Yes, we need to supply data to GPU servers [and] it’s a relatively straightforward product extension.”

WekaIO and Spectrum Scale

We ask if WekaIO is a competitor, to which Shea responds: “In general, no. We have different customers.”

A typical Panasas customer has tens to hundreds of different workloads and 2 to 20 petabytes of data, he says. They can throw any and all of it at a Panasas’ filesystem because it auto tunes, and copes with small files, large files, medium files, heavy read loads, write loads or mixed IO loads, whether for HPC or AI jobs, he adds.

And IBM’s Spectrum Scale? It is “a worthy competitor and probably our closest competitor.” He adds that he thinks it is quite complex, however.

What about containers? “We see it as a tool customers are using. I think that it will become a big deal as more HPC is done in the cloud.”

Public cloud elephant

Yes, the public cloud, the elephant in the HPC room.
Shea admits: “Customers like the hybrid cloud. They want to burst to the public cloud.”

He tells us he wants to extend Panasas’ value proposition to the public cloud in a way that makes sense.

The thumping great problem is moving data when you have files in the terabytes and petabytes of data. “How do I get the data where it needs to be?” says Shea of the issue.

He believes the NetApp direct access model is plan A for many vendors, in which arrays are placed in locations with fast, direct access to the public cloud, so that cloud compute can access it at low latency. But he says: “Our customers are concerned about public cloud lock-in.” It’s a big HPC challenge. 

Panasas could theoretically move its PanFS filesystem software to the public cloud “but it shouldn’t run on standard AWS or Azure. … it needs to be part of the underlying cloud infrastructure.”

Shea says it’s a long game and that he thinks all of the public cloud vendors are looking to HPC for growth.

How about the shorter term? Shea thinks life sciences customers need to start at the half-petabyte scale and Panasas’ ActiveStor Ultra minimum is a petabyte. 

He says: “We’re working on an all-flash product in smaller configurations.”

It will probably use TLC (3bits/cell) flash and not QLC (4bits/cell), which is slower and has less endurance than TLC. There might be room for QLC in the future, as an additional storage tier.

Comment

Shea is a continuity CEO, in the sense of taking the existing organisation and product range and incrementally developing it. Panasas’ technology is good and has unique advantages, such as the auto-tuning. It has already spent some time in the traditional HPC market, and as HPC techniques are being used in the enterprise AI and machine learning area, that forms a natural extension of Panasas’ HPC market.

Partnering with other suppliers in the enterprise AI/ML makes obvious sense as does a focus on hybrid cloud where the customer is looking to avoid public cloud vendor lock-in. Dealing with the data gravity problem inherent in such a hybrid cloud approach and devising a way to partner with public cloud vendors will be two large and complicated issues in Shea’s in-basket.