A Quick Note on The Similarities Between Servers and Aquariums
(No, I am not crazy; there is a point)
Making a server is like buying a fish tank. When you walk into the pet store, you feel good. All you need is a tank, some fish, and some water (which you already have). Easy. But then, you meet the aquarium expert. You realize owning a fish tank is not just throwing water and fish into a glass container. You must buy filters, heaters, food (you can’t believe you forgot about that), and all the other paraphernalia associated with owning an aquarium. So, you do that, dismayed at the expanding costs, and go home to set up your aquarium. The following week, everything is ready. You go to the fish store and ask for fifteen fish. Looking sympathetically at you, the aquarium expert says, “Sorry, the aquarium unit you bought can only support two fish.” Owning a fish tank is a challenge.
Similarly, when you have a server, the costs, space, and frustrations are not only from the acquisition costs to buy the drives but also the infrastructure costs necessary to support the drives. And let us remember to calculate all the energy this will consume. So, after you build a server, even if you know what you are getting into, you still feel incredibly frustrated, just like when buying a fish tank. “I’m paying this much, and I only get that!” Disappointment would be an understatement. However, do not despair; there is a solution. If I told you that you could significantly reduce the cost of an aquarium by simply changing the fish inside? You might think I was crazy. You would be right. Suppose I told you that you could significantly reduce server expenses by simply changing the solid-state drives that you were using. In that case, you might again think I was crazy. However, this time, you would be wrong. By using ScaleFlux solid-state drives with compression, you can significantly reduce energy and cooling costs/infrastructure while at the same time considerably increasing drive density and drive speed. Owning a server may be easier than owning an aquarium.
Energy and Cooling: Laugh at Anyone with A High Server Maintenance Bill
Don’t laugh at them. They already know it’s too high, and rubbing it in would be cruel – but it’s true. Because using ScaleFlux drives with compression significantly saves you money on cooling and energy costs.
Let us show this to you in practice, objectively proving our drive’s tangible benefits. To do this, we ran the same random write test on a preconditioned 7.68TB ScaleFlux drive and a preconditioned 7.68TB competitor drive, measuring the power from each drive. Both drives were running at about 200K IOPS. Look at the results.
That’s a competitor drive – Boo!
You can see that at 200K IOPS, the drive consistently uses an average of 17 Watts. Not bad, especially considering that this drive was running the write test after preconditioning, as it also had to perform garbage collection. Now look at our drive’s power measurements.
At first glance, the competitor’s drive could appear to be a good option. Still, at 200K IOPS, the ScaleFlux drive uses about 13 Watts on average, four fewer Watts than the competition. Now when you reduce energy consumption by 4 Watts for each drive across the entire data center? It starts to have a meaningful impact on the operational costs of your storage infrastructure.
However, this is only part of the picture. In this test, our drive was limited to 200K IOPS, meaning it could not run at full speed. Therefore, the energy savings were less pronounced. How would the IOPS compare if our drive could use full power? Spoiler: our drive crushes the competition. When both drives were allowed to run as much energy as they wanted during the same random write test, they used similar amounts, around 16-17 Watts (if you’re curious, we were the drive closer to 16 Watts).
When tested like this, the competitor drive got around 212K IOPS – not bad. On the other hand, when tested in the same way, our drive achieved about 709K IOPS. By preconditioning both drives similarly, our drive got almost 3.5 times more IOPS per Watt. Here is a more visual presentation of this:
When you use the ScaleFlux drive, you can realize massive power savings. It might be clearer if we had a fish for scale.
Let’s make an example of this claim to put this into perspective. Suppose you had to buy eight drives to meet a specific target. Each of these drives uses about 16 Watts of power. That is a total of 128 Watts that you are using to make your target. If you used ScaleFlux drives, you would only have to use 2-3 drives, and those few drives would use only 32-48 Watts of power. You could save almost 100 Watts of power per node.
We see this also translate into massive cooling benefits. When a drive uses energy, it gives off heat. The power a drive uses directly corresponds to the heat it gives off, which directly corresponds to the cooling infrastructure you must install. Since ScaleFlux drives use significantly less power than competitors’ drives, when using ScaleFlux drives, you do not need as much cooling. You can therefore save enormously on the cost of cooling. In the end, getting a ScaleFlux drive is a win-win situation: you save on power and cooling. Or it could be a win-lose situation because you’ll have to exercise restraint and not laugh at anyone else’s high server maintenance bills. I’m still determining, but you decide for yourself.
Drive Density and Speed: Up, Up, And Away!
I wish that I could put more fish into my aquarium. I can’t. If you want to, you could fit more drives into each server. I know life is unfair, but to quote Calvin and Hobbes, “Why isn’t it ever unfair in my favor?”. Anyway, when using ScaleFlux drives, you can increase drive density within your servers. How, might you ask? Magic!
Here is how our magic works. Each server rack in a data center can only give the drives a certain amount of power. That power is finite so that it can power x amount of competitor drives. However, since ScaleFlux drives can use about 3.5 times less energy than the competitor drives, you can fit about 3.5x ScaleFlux drives into that server rack. That means you can squeeze about 3.5 times more drives into the data center than alternative competitor drives. Just as I said, magic!
ScaleFlux drives also have much less latency than competitor drives. Looking at the same random write test where we compared power metrics, we compared latency between the two drives. The results are below.
No explanation is needed; ScaleFlux drives deliver much more performance than competitor drives!
Concluding Thoughts: Comparing Fish Tanks with Servers
For those in the know, when you buy a fish tank, common wisdom suggests that for each fish-inch, 1-gallon of water be provisioned (FInGals). Therefore, I could easily overspend on a five-gallon aquarium, supporting few fish yet incurring high costs — a disappointment, to be sure. Servers are the same, spending too much and getting too little while incurring the infrastructure overhead costs of a much larger deployment. It doesn’t have to be like this; with ScaleFlux drives, servers can be more efficient, saving energy and cooling, all while increasing drive density and speed.
The moral of this story is: avoid deploying your servers like a fish tank; use ScaleFlux drives.
