First, I want to take a few moments to describe how at Sovereign Systems we go about testing all flash arrays.
You probably are aware by now the major difference that flash storage introduced in how data is written and maintained inside a SSD.
In comparison, a spinning drive is bit level addressable – meaning I can change a single 1 or 0 to something else, flash is page level addressable – meaning I have to write an entire page of storage even to change 1 bit in the data. On top of this you have maintain the flash cells life because of how many times you can rewrite a flash cell.
This whole maintenance process is garbage collection. So we are very aware of this in our testing in that when we get a flash array we want the garbage collection process running like the array were well in to its production use, not like it was fresh out-of-the-box.
Here’s an analogy of how we do this: there is a guy that can drink at most 10 beers a night before he passes out, we’ll say his name is Ian. So normally Ian tries to drink 8 beers a night because he knows he can still function effectively and call an uber to get him home.
One night he goes to a different bar where they are known to rob you when you pass out. The bartender knows Ian isn’t a regular and makes him drink 12 beers. Just as expected, Ian passes out and it takes a little while for his body to filter out the excess beer before he can sober up enough to call an uber. But uber won’t pick him up because all of his credit cards have been maxed out.
That’s a bad night for Ian, but what we know that is that at 80% of his capacity, his internal organs are working in right order and his brain is still working (if you ever meet Ian you’ll be impressed).
There are a few other tests we will run after Ian has had his 8 beers, like can he stand up if we kick one of the bar stool legs out from under him. Now let’s talk about less about beer and more about storage.
We have been supporting storage solutions long enough where we’ve seen a few really bad situations and we’ll try to recreate those situations in our tests and see how these systems respond.
Let’s talk about the array that we recently tested that surprised us with its results.
It’s the Kaminaro K2.
Performance is usually the first topic on most people’s mind in the AFA space. In our testing we were able to achieve, in a single Kblock, approximately 225,000 IOPS at an average response time right at 1ms. Most response times were below 1ms but with larger IO sizes we expect long response times, which did bring up the average. For this specific test we used vdbench to generate a workload across 8 CentOS virtual machines. We fixed the requests to 8kb IOs at 50/50 read write ratio.
Next, availability. Five nines are tables stakes now in the AFA space and we are not even going to evaluate something that does not play at that table. We attempted a few horror tests on Kaminario with production systems on the array. All of them successful, including a non-disruptive upgrade that was really not disruptive to our normal operations. We normally don’t drive as much IO as we tested, but we saw no degradation of our normal performance during the upgrade and no loss of access to the storage. Things just worked like they were supposed to. Oh, we also did the upgrade while our SovLabs were building a few new modules to deliver to one of customers. They did not see a blip.
Now, the architecture. The next generation data center needs options, and we are able to deliver that with Kaminario. With an everything active architecture we simplify the model and then the ability to scale up and out is great for controlling and balancing the costs of growth. Most of the time our customers need just a little more capacity, but sometimes you’ll need the horse power that comes with more controllers and it’s great to have both options with Kaminario.
And then support. Our interactions with the support team has all been positive and easy. Meaning it does not take 5 emails, 12 support tickets, 8 conference calls, and threats to get something done. Most of the time the Kaminario team is reaching out to us before we know something has happened. And unfortunately in Atlanta during the summer months everyone kicks up the A/C causing frequent brown outs in our data center, so we live with a couple outages each summer. Each time Kaminario has been first to alert us to an issue. And the array keeps running along.
Finally, what we like from Kaminario is a vision for the all flash data center. From what we have seen from the team, and are eagerly waiting to see more, it looks like a bright future for staying on top of the new technology curve that is enabled by all things flash.