Cheeky upstart Mangstor hungrily eyes Fusion-io's pots of gold

Some seriously supercharged speedy server SAN

Server SAN

It is the NMX6320 which has nodes with MX6320 cards and which scale-up from 8TB through 16 to 32TB. This SAN, Mangstor says, “easily scales from 10s to 100s of TBs", by adding nodes in what looks like a cluster.

The starting node and scaled-out nodes’ MX6300 cards are made available to accessing application servers as a single virtual SAN courtesy of a Titan software stack.

The NMX6320 nodes are interconnected with an RDMA fabric using RoCE, iWARP or InfiniBand protocols and 10, 25, 40, 56, or 100Gbit/s RDMA network links as appropriate.

Accessing application servers use similar networking infrastructure to have NVMe read and write access to this server SAN “nearly identical latencies as if they were accessing local PCIe SSDs; increasing application performance and providing over 3.5 million random 4K read and 2.25 million 4KB write IOPS.”

In effect, we have an NVMe fabric in operation with networked storage accessed at local PCIe SSD speed; amazing, and of course you pay for the privilege.

Here are Mangstore’s numbers:

• Read/write latency is 90 µs/15 µs
• 8TB — 1.75m/1.10m R/W IOPS, 7.0/4.5GB/sec sequential R/W bandwidth
• 16TB — 3.5m/2.25m R/W IOPS, 14.0/9.0GB/sec sequential R/W bandwidth
• 32TB — 3.5m/2.25m R/W IOPS, 14.0/9.0GB/sec sequential R/W bandwidth

It makes other server SAN and virtual SAN products found in hyper-converged appliances, such as those from Nutanix, Simplivity and the EVO: RAIL crew look pedestrian in comparison.

The software integrates with OpenStack Cinder and there is a REST API, for, Mangstor says “integration into 3rd party cloud management systems". It provides native per-volume replication for backup and supports site-to-site disaster recovery.

Company background

Mangstor is based in Austin, Texas, and was started up in 2011, becoming incorporated in March 2012 and picking up $4m in A-round funding in 2013. That doesn't seem a lot of money for developing proprietary PCIe card hardware, firmware and software.

We'd think a second funding round will be sought in the next few months, if it isn't ongoing already, to fund product development and build out the company’s business infrastructure.

The founder is Ashwin Kamath who was an engineering director at Emulex. He is Mangstor’s chief development officer, with Trevor Smith, an ex-engineering VP at Emulex, being the CEO. Paul Prince is the CTO, and he comes from Dell where he held a CTO position.

Mangstor looks like it has a very hot PCIe flash card and a blindingly fast server SAN. The marketing proposition will for sure focus on the performance advantage. It should give Mangstor a fairly clear run at performance-limited applications where customers need performance so much that they will take a punt on a startup supplier.

However, EMC’s acquired and developing DSSD technology looks to be aimed at the same marketing ballpark; scale-out, all-flash arrays with fast networked access.

It seems obvious to us that Mangstor will need more funding to keep its development on track and comparable to (or ahead of) EMC’s well-funded, we assume, DSSD tech.

Huawei’s Tecal ES3000 could also be developed to provide more speed although Huawei developing server SAN software might be a bit of a stretch. It shouldn’t be ruled out though.

El Reg is not aware of any other equivalent technology from other suppliers. If the market for such products becomes substantial then Dell, HDS, HP and IBM will potentially look to acquire the technology and products needed to enter it.

This makes Mangstor an attractive acquisition target, if it manages to deliver on the promise of the flashy product technology it is developing. ®

Bootnote

Can’t say we’re that smitten with the name Mangstor. The “stor” part clearly refers to storage but what the heck does “Mang” refer to? Perhaps it’s best not to ask.