SolidFire Aims SSD System At Cloud Providers

With the number of acquisitions in the storage business lately, you’d expect that we were in a mature market with declining product differentiation and innovation, but, from where I sit, nothing could be further from the truth. As fast as Dell and EMC can hoover them up, startup vendors keep coming up with new ideas and solutions aimed at the specific needs of targeted markets. SolidFire’s new all-SSD, scale-out, 10-Gbps iSCSI array that’s targeted directly at the cloud storage provider market is a case in point.

Since the SolidFire Storage Solution was designed for the cloud provider market, it came as no surprise to me that its design and feature set fit that user community’s needs especially well. To start, since the SolidFire system is a scale-out storage system based on industry-standard 1U servers and solid-state drives (SSDs), providers can buy capacity--and, of course, performance--as they add customers. A single system can have as many as 100 SF3010 storage nodes, each of which has 10 300-GByte SSDs, and the cluster can support as many as 100,000 volumes.

While enterprise users standardized long ago on highly reliable storage systems using multicontroller NAS systems and Fibre Channel SANs, those solutions have been too expensive for most cloud providers. Instead, they’ve built clustered file systems and object stores that scale out. It will be easier to insert SolidFire’s iSCSI solution into those data centers than competing solutions from Texas Memory Systems or Violin Memory that are Fibre Channel-based. Being based on iSCSI also simplifies scale out, allowing SolidFire to use IP redirects to send I/O requests to a node that holds the data chunk being accessed.

I was impressed by how SolidFire takes advantage of the dual six-core processors in each node to do inline deduplication and compression to squeeze more data into the expensive SSDs. As data is written to the system, it’s broken into 4K chunks and hashed; data chunks are compressed and stored in what is essentially content-addressable storage where data chunks are stored by their hashes; and a volume store tracks chunk locations and maps chunks to user volumes and blocks.

Of course, whether users will really get 10-TBytes of usable space out of the 3 TBytes of SSD in a node depends on how well their data will deduplicate and compress. Some cloud provider data, like virtual machine system drives, will deduplicate very well. MySQL databases will compress pretty well, so SolidFire’s projection might be on the money. It will depend on the user’s data. SolidFire will distribute a free data scanner that prospective customers can run against their data to see how well SolidFire’s techniques will reduce it.Rather than use conventional RAID across the 10 SSDs in each enclosure, SolidFire mirrors each data chunk across two nodes in the cluster. They promise to support three-way mirroring in a subsequent release.

The SolidFire OS also includes multi-tenancy features including LUN by LUN quality of serviceincluding minimum and maximum IOPs and both IOP and throughput burst control. This allows service providers to ensure that one customer’s IO sucking application doesn’t slow everyone else down.

When asked about pricing, SolidFire would only say that the system would be cost-competitive with a spinning disk system. Since that spinning disk system could be anything from a Clariion with equivalent capacity on 300-GByte 15K RPM drives to a universal storage platform (USP) with hundreds of short-stroked drives, "cost comparable to spinning disk" sounds like it might be a bit expensive to me.

All in all, I’m impressed by the SolidFire system. Scale-out, 10-Gbps iSCSI with real multitenancy support looks like just what cloud providers should be looking for to build their equivalent of Amazon’s EBS (Elastic Block Storage).

At the time of publication, SolidFire is not a client of and has no business relationship with Howard Marks.