There's a widespread mistaken belief that high sequential speeds are what matters. Truth is they don't except in certain very esoteric applications (real-time video editing, copying large video files from one SSD to another SSD). It's actually the small file speeds which make the biggest difference.
The problem is that humans don't perceive storage speed in MB/s. We perceive it in how much time we have to wait for an operation to complete. That's sec/MB, or the inverse of MB/s. Since it's an inverse, the smaller MB/s figures end up having the bigger impact. If asked which would result in a bigger speed improvement - upgrading from a 125 MB/s HDD to a 250 MB/s SSD on a SATA 2 port, or upgrading from a 500 MB/s SATA 3 SSD to a 3000 MB/s NVMe SSD, most people would say the latter. After all, the former is only a +125 MB/s improvement, while the latter is a massive +2500 MB/s improvement.
But do the math. Say you need to read 1 GB of sequential data from each of these drives.
125 MB/s HDD = 8 sec
250 MB/s SSD = 4 sec, a 4 second reduction in time spent waiting
500 MB/s SSD = 2 sec
3000 MB/s SSD = 0.33 sec, a 1.67 second reduction in time spent waiting
So because the scale is inverted, it becomes non-linear. And the +125 MB/s improvement, by virtue of being closer to 0, actually reduces time spent waiting by more seconds than the +2500 MB/s improvement.
The same goes for sequential speeds vs small file speeds. Because the sequential speeds are so much faster in MB/s, they have less impact on overall wait times. The slower small file read/write speeds end up dominating performance measurements. e.g. Say you needed to read 1 GB of sequential data and 200 MB of small file (4k) daya. Which is faster - a SATA SSD with 500 MB/s sequential speeds and 50 MB/s 4k read speeds, or a NVME SSD with 3000 MB/s sequential speeds and 30 MB/s 4k read speeds?
(1 GB)/(500 MB/s) + (200MB)/(50 MB/s) = 2 sec + 4 sec = 6 sec
(1 GB)/(3000 MB/s) + (200 MB)/30 MB/s) = 0.33 sec + 6.67 sec = 7 sec
Surprise! The older tech drive is faster. Again, because the scale is inverted, the MB/s which is closer to zero (50 MB/s vs 30 MB/s) ends up dominating the comparison, and the huge MB/s numbers (500 MB/s vs 3000 MB/s) have a much smaller impact in comparison. Since the older drive had better 4k performance, it ends up faster overall.
Which brings us to RAID 0. RAID 0 works by dividing files up into two pieces, and sending each drive one piece to write. The same process happens in reverse for reads. This works for large files, especially on HDDs. But for small files... the smallest sector size is 4 kB. If you try to write a 4 kB file to RAID 0, it breaks it into two 2 kB pieces, and pads both out with zeros to make them 4 kB. Then sends each drive a 4 kB file to write. Exactly the same as if you weren't using RAID 0.
Once you add in overhead for all this splitting and padding, RAID 0 actually ends up being slower than a single drive at small file (4k) read/write operations. And since it's the smaller MB/s speeds which dominate wait times the most, there ends up being very little benefit to RAID 0 on SSDs and even a possible performance decrease, outside of those very esoteric applications (real-time video editing, copying large video files from one SSD to another).
The same problem crops up with car fuel mileage. For some reason the U.S. uses MPG, which is actually the inverse of fuel mileage (the rest of the world uses liters per 100 km which is the proper way to measure mileage - fuel consumed divided by distance traveled). Because MPG is an inverse, a constant delta in MPG represents larger fuel savings the closer it is to zero. And switching from a 15 MPG SUV to a 19 MPG SUV (a "mere" +4 MPG) actually saves more fuel than switching from a 30 MPG econobox to a 50 MPG hybrid (a "massive" +20 MPG). If you drove each car on a 100 mile trip:
15 MPG SUV = 6.67 gallons
19 MPG SUV = 5.26 gallons, or 1.4 gallons saved
30 MPG car = 3.33 gallons
50 MPG hybrid = 2.00 gallons, or 1.33 gallons saved
Facebook
Twitter
Instagram