When I think about what matters in OLTP, the first thing that leaps to my head (and maybe some of yours too) is performance – raw performance. There’s nothing like super-speedy Orders Per Minute (OPM) to get some attention. The faster they go, the more we get, the better.
Then I take a step back and think about what else matters. We want great responsiveness (low average latency) and we want it to be consistent (say, 99.9th percentile latency). Those are both good too. But is that it? With the pervasiveness of flash, is there something more we should expect?
We should expect efficiency. Tons of OPM, low and consistent latency are great – but what do we have to do to get them? We should think about power efficiency (at the system level, not the drive level – something like OPM/watt).
We did just that. We tested what us older folks used to think of a ‘performance’ (16 high RPM hard disk drives spinning about, trying hard to keep up) against four and eight of our 5100 MAX SSDs. The results are not for the faint of heart.
Quick summary: the 5100 MAX configurations bested the legacy option in every metric we tested – OPM, average latency, latency consistency and system-level power efficiency. That’s it, end of summary.
Interested in more details? Then read on...
We want to be fair here – you’ll notice that the 3 configurations we tested don’t have the exact same capacity. We used a small SSD setup (4x 5100 MAX, 1.9TB total capacity), a medium HDD setup (16x 15K SAS drives, 2.4TB total capacity) and a large SSD setup (8x 5100 MAX, 3.8TB total capacity).
Was it a tough battle? If you are an HDD fan, yes. If you like the flashy contestant, not so much.
Orders Per Minute: Flashy OLTP – 3.8 to 10.7 times Better
The 5100 MAX is a practical, cost-optimized SSD that enables phenomenal OPM compared to the legacy stalwart (the 15K HDDs). This ‘performance’ class HDD setup is painfully slow – a fact that became very clear as we tested.
The HDD setup got to almost 2,600 OPM (not quite, but close enough). Our small SSD setup, nearly 10,000 (about 3.8x more). Our large SSD setup up cranked out nearly 28,000 OPM (about 10.7x more).
Hmmm…seems to confirm what we expect from flash. Flash wins round 1.
Responsiveness and Consistency: Flashy OLTP wins again!
We also measured responsiveness (average latency) and consistency (99.9th percentile latency) for the same three setups. Flash won this one as well – 33% to 41% lower average latency that’s 60% to 73% more consistent. Round 2 goes to flash.
What about power efficiency? Do the spinners win here? No.
Finally, we measured system level power consumption while testing OPM and calculated power efficiency (by dividing each configuration’s OPM by the measured power consumption) for each of the contestants. If we consider the HDD configuration OPM/watt as our baseline, we see that the 4x 5100 MAX is 4.7x more power efficient than the HDD configuration [(33.9 OPM/watt) / (7.3 OPM/watt)]. The 8x 5100 MAX configuration is 13.5x more power efficient [(98.2 OPM/watt) / (7.3 OPM/watt)].
Yeah….seems our spinners came up short. Again.
So? What’s the surprise here?
I doubt any of you really expected the spinners to win the basic performance (OPM) test. I’d also bet that few (if any?) of you expected them to have lower latency. I’d venture that maybe some of you thought their latency consistency might come close. Any takers?
And power efficiency? That may have been a surprise to some, maybe not to others.
Looking at storage more holistically – we’re striving for increased OPM, better and more consistent latency as well as improved power efficiency. On these counts, the 5100 MAX delivers.
Want to learn more?
Take a look at the complete Technical Brief (with more details on configuration, testing, results and comparisons) and the complete 5100 lineup (MAX, PRO and MAX families in both 2.5” and M.2 form factors). If you have questions about the testing we ran, tweet me @GreyHairStorage or connect with Micron on Twitter @MicronStorage and on LinkedIn.