Tech Field Day can be an interesting event, not just from the technology side. One of the early questions for Violin Memory was basically: How are you still in business? The answer was that they still have cash and are selling product. The core of their presentation is that this is not the Violin you remember from a couple of years ago. So I’d suggest you take a moment to let go of what you know about Violin. Forget that they were a one product company. Forget that they had a speed demon all flash array with no data services. Forget that their IPO was a disaster. How are you doing? All forgotten? Let’s take a look at the new Violin.
Today’s Violin wants to be a leader in the modern All Flash Array (AFA) market. Naturally they have deduplication and compression options to get the most data into the array. They also have storage snapshots as well as both sync and async replication. They even have an agent to go inside your VMs and physical servers to make the snapshots application consistent. They provide both Fibre Channel and iSCSI connectivity to the array. Also like other AFA vendors, their array is built around standard x86 servers running smart software. So far they seem to have table stakes features for an enterprise AFA. The secret sauce for Violin is that they don’t treat flash at all like a disk. Rather than having flash in SSD form factor with a SAS or SATA interface they have their own physical form factor. Violin have custom designed a Violin Intelligent Memory Module (VIMM) that takes the role of an SSD. Violin also have their custom designed RAID controllers which are not designed to talk to hard disks or SSDs but only to VIMMs.
There is a very strong argument that getting the most out of solid state storage requires that you not treat it like a hard disk. SSDs and the protocols that are used to access them are legacy hard disk constructs. SAS and SATA protocols were designed for hard disks. Inside an SSD there is a translation layer that handles making flash look like a hard disk and in doing so reduce performance. Most array vendors use these legacy items because the engineering is already mostly done. An SSD is so outstandingly faster than a hard disk that getting 30% of it’s performance is still a huge benefit. This is the approach for price driven AFAs and hybrid arrays which mix SSD and hard disk. Violin has taken a different approach and aims to squeeze every last bit of performance by doing a lot more engineering in hardware. The VIMM is a custom carrier with interconnects for flash chips. The RAID controllers are setup in an HA configuration using a RAID format optimized for flash. This additional engineering is to deliver the maximum performance from the flash chips. Using a CPU for deduplication and compression does add latency to transactions. So Violin allow you to turn off deduplication and compression on a per LUN basis. For highly latency sensitive workloads, losing data efficiency to get better performance may well be worthwhile.
Violin Memory is trying to come out of it’s tumultuous last couple of years. It will be interesting to see whether the market has faith in the new Violin. The products look good and the people we met were all open and clear in their communication. I hope they do well.
Disclosure: I attended TFDx at VMworld USA and was provided a nice lunch and an uncomfortable chair. Violin also gave all the delegates a Grateful Dead themed T-shirt that was rejected by my teenage daughters. Neither TFDx nor Violin requested or reviewed this blog post.
© 2015, Alastair. All rights reserved.