StorONE’s new S1 array provides astounding storage speed in terms of IOPS and latency. But it does not yet use NVMe drives nor NVMe-over Fabrics (NVMe-oF) networking access.
We think that both seem logical and asked StorONE CEO and co-founder Gal Naor about this.
B&F: StorONE tested its S1 system using HGST SSDs inside a Western Digital 2U24 chassis and found it delivered delivered 1.7 million random read 4K IOPS at less than 0.3 ms latency. These were SAS-connected SSDs. Can you go faster?
Gal Naor: When HGST did the testing, we used standard off the shelf SSDs. We can also use NVMe and the numbers are much higher and the latency is even lower than our initial testing. We have done testing already with a single Intel NVMe and will share the numbers shortly (we managed to achieve ~350,000 IOPS from a single NVMe drive, more than 1.5M IOPS with 4 NVMe).
B&F: Is StorONE considering implementing NVME-oF access to its array? It seems logical that it would.
Gal Naor: NVMe-oF – I totally agree with you and this is an approach we are pursuing and the support is in the works with relevant partners.
B&F: We have learnt that NetApp’s coming MAX Data product will cache or tier back-end NetApp all-flash array data in persistent memory (Optane DIMMs) in accessing client servers. These client servers will use an NVMe-oF link to the NetApp array and also have Plexistor SW that presents the Optane DIMMs as a POSIX-access block store to applications. The Plexistor SW by-passes the client servers operating system I/O stack and will be able to offer, it seems, 4 – 5 microsecond access latency to data in the Optane DIMMs and, consequently, have a high IOPS rating.
How would StorONE view such an offering from NetApp and position its own offering in relation to it?
Gal Naor: This is another way to demonstrate the uniqueness, beauty and benefits of our technology. We manage to achieve same results with SW while NetApp needs to add more layers and complexity and throw more hardware at the problem. Our advantage will be in CAPEX, less HW to buy and with the OPEX there are less components to manage, to integrate, less moving parts and less finger pointing in case of performance problems.
They are NetApp so the customers are used to spending exorbitant amounts of money to continue to stay in their ecosystem. You can get a system from StorONE that will accomplish the above for less than what you would pay for their service contracts alone.
B&F: You would argue that storage stack software inefficiency causes a lot of latency?
Gal Naor: A few customers at VMworld told us that comparing StorONE to its competitors is like the competition between old flip phones/analog vs. smartphones. Before the smartphone, all the functionalities existed, cameras, GPS, telephone, internet, calendar and notes. The smartphone managed to solve all of these challenges by transfer them to software and not by adding more complexity and more HW to take with you. These other solutions continue to use HW since they are making up for the inefficiency of their software. With us you don’t need to spend money to try and make up for software issues that cause latency, etc.
B&F: Are you saying the hardware speed acceleration route costs more money than doing software improvements?
Gal Naor: We are … talking about systems that are priced much higher than ours and if we put a few systems together we will be faster, provide complete enterprise storage functions that are running at all times, and will be pennies on the dollar for what others will charge.
We know that achieving any speed or capacity these days is possible if you throw enough money at it, but is this really what companies are looking to do? With StorONE you are doing it with a smaller footprint, much better pricing and of course complete data protection. Nobody can do this but StorONE.
B&F: This StorONE software is actually unique. No other supplier we know of can place ‘its storage controller’ software in front of drives and get this level of performance. It has us a tad concerned; if something looks too good to be true then it probably/possibly/might be too good to be true.
And yet, test runs are test runs. We eagerly await independent verification.
In the meantime, check out a StorONE test run video:
As an afterthought, we wonder if it shares any similarities with Solarflare’s kernel bypass technology, although that can’t be the whole story by any means. B&F
