Drive speed. 5400 or 7200 for 3.5 inch

I’m buying a larger drive for one of my storage nodes and not sure if buying the faster 7200 speed will offer any benefits. The pro’s are that they are faster, but the cons are:

  1. use more power
  2. are louder
  3. Cost more
  4. run hotter so may need cooling depending on environment (and it can get to 40C here).

So is the faster speed and lower latency worth it or worth anything at all?

Just make sure that it is not a SMR.
In other cases, variations in seek speed may have impact on ability to win the race for pieces.

1 Like

There are (almost?) no true 5400 drives produced anymore. The drives marketed as 5400 are actually 7200 drives that are slown down at the interface level to imitate slower drives and discourage their use for more professional applications. This is easy to check using a tool for measuring acoustic profile. See this article as an example.

Drives do differ in terms of noise and temperature, but rotational speed is no longer a factor. Instead, look for helium drives or drives with a smaller number of platters.

I use 7200RPM drives for my node. One reason is that they would have slightly lower latency than 5400RPM, but another reason is that I bought proper server-grade drives (Ultrastart HC330) and most, if not all server-grade drives are 7200RPM or more.

This is interesting, I do not have any new 5400RPM drives or I would check them with an oscilloscope (motor drive frequency).

My drives tend to be used enterprise drives, because they are cheaper. and they are almost universally 7200rpm.

And as already mentioned, most large capacity drives are 7200 rpm nowadays. and many marketed as “5400 rpm” are slowed down in firmware, but inherently rotate at the faster speed and so don’t help you with noise or heat.

the few larger capacity drives advertised at lower speed (I’m thinking seagates advertised around 5100rpm) are more likely to be SMR, which should be avoided.

But after all that, if you truly find two similar CMR drives, one 5400 rpm and one 7200, then the difference will be “not big”. the peak sequential speed will be 25% lower, which won’t matter, and the access time will be a few ms longer. But that’s just a couple ms on top of the inherent much longer latency imposed by the internet

1 Like

I am not sure whether to believe such a claim.

Is there evidence supporting the hypothesis that variations in HDD access latency (~5 milliseconds) have measurable effect on Storj egress or ingress?

2nd question: Is there a difference in Storj egress, or ingress, between a 2TB HDD and a 2TB SSD? There are two cases to consider: (a) when each drive has 1TB of free space available, and (b) when both drives are full.

1 Like

I feel like there is no real benefit to have top notch hardware for Storj… Personnally, I prefer a cheaper solution (cost and power wise).

Besides, like @atomsymbol & @EasyRhino I’m not sure gaining a few milliseconds will radically change the number of successful egress requests, on the grand scheme of things…

I mean in absolute terms @Alexey is right. But maybe it doesn’t impact the node that much ^^

But yeah, no SMR whatsoever if possible.

Some data:

  1. I have 2 nodes.
  2. The 2nd has a significantly higher network latency than the 1st.
  3. Contrary to point 2, yesterday the 2nd node has uploaded 10.2GB (egress) of data and the 1st has uploaded 8.7GB of data.

There is a strong relationship between [ingress over the past few hours] and egress. Approximately 50% of egress depends on what the ingress in the past few hours was. The egress-ingress dependency has a much higher impact on node performance than the differences between HDD and SSD latencies as well as the differences in network latencies.

3 Likes

There are so many variables in the race for pieces and it’s likely drive response is one of them. I know my 4TB drives slow down when full and I’m up grading to 8TB drives. I will get 7200 rpm drives just to counter any more slow response due to the extra capacity. They also come with a longer warranty.

Also, if egress goes up, slower drives may not be able to keep up. I use SLOG for writes, but reads would still be from the HDDs (unless the traffic was a few files being accessed a lot instead of lots of files being accessed once or twice).