I’m thinking about replacing my 2x3TB drives to something new and bigger.
I’ve stopped my choice on Seagate Exos X12/X14 12TB.
But, what I want to ask is - given the high intensity writing/reading, which is the best drive in 8-12 TB category? The most durable, the fastest and so on.
Budget is not limited.
Thanks in advance
Personally I would go for Seagate Exos 12TB/14TB as it has given me zero issues in 3 years. Also its meant for 24/7 usage and works great with random read/writes which Storj needs. You would also have to make sure you have a good CPU too
iowait of any kind
Since budget is not an issue I would advise to wait for Seagate’s 20TB HAMR drive coming in 2021 (tentatively).
Thanks a lot for your reply!
I forgot one important detail - my current drives look like they’re failing. From time to time they go to 90%+ activity with small write speeds and high response time (although this doesn’t prevent me from reading/writing other data at normal speeds).
So, probably a good idea to replace them now, before they fail.
As for CPU, I have octa-core AMD FX8350, it’s old, but it does it’s job.
That happens due to the storagenode checking its files periodically. Does SMART tell you the drives are failing?
You will see 100% activity too when your node grows due to directory traversal. Also have good amount of ram as you use it for other purpose too.
What ever you choose, stay away from SMR drives.
I’m using two SATA-attached Western Digital Gold 12tb since more than one year for Storage Nodes and my own data on Windows 10, but its obviously too early for having issues at this time. Though, people recommended to check the reports made by Backblaze in their datacenters; https://www.backblaze.com/blog/backblaze-hard-drive-stats-q2-2020/
As you can see, Seagate drives still have more crashes in their experience than WD/HGST. I also took that in consideration when I bought my drives. I’ve now two HGST 18tb but I didn’t still installed them. WD/HGST are typically more expensive, and I read the cache function give a better write speed for Seagate, and WD/HGST are also noiser, but its not really a critical point for the activity of the nodes.
Basically, any entreprise grade hard drive is good for storage nodes (except SMRs as mentionned), since the most come also with 5-years warrantly. It worths the value.
Well, no, not really, they pass SMART tests for now, but I can see relocated sectors count 100 and some other errors in the SMART data. So I’d rather replace them before it’s too late
Thanks for the advice, I’ve got 16 GB of RAM, I guess, it’s more than enough
Thanks for the advice, I haven’t heard about Seagate having more errors and fails.
I’ll take a look at WD in this case, can you recommend which drives from WD should I check? Also, how do I differentiate SMR drives? I’ve heard that WD are hiding this fact
You can check this thread:
the best hdd’s for running a storagenodes… is more than one…
run 2 or 3 nodes of similar sizes and you will have 1/2 or 1/3 the load on either drive, most single drives storagenode will at times have difficulty keeping up, without some sort of modern caching solution.
similar sizes aren’t really required, it just helps the load balancing a lot… but really running a single node is usually a mistake… i would the the minimum should be 2… simply because it will increase the lifetime of the hdd’s running the storagenodes.
and ofc 1 hdd for each node …
I very much doubt that is true. If this were the case we would see many more complaints on the forums about this. But in fact the complaints that we did see were usually either SMR drives, really old ones or connected through slow interfaces.
It’s not a bad idea to run more than one to balance it out, but I don’t want to give the impression that you would get into trouble otherwise. Many SNOs are doing just fine on a single HDD.
sure one can run single drive storagenodes and many do.
and adding a second node can always be done later…
the limitation is almost always the iops, when it comes to performance… ofc an argument could be made for going towards the largest possible drives to save on electricity costs… ofc a regular hdd wouldn’t be able to use it’s cost in electricity in less than like 10 years or whatever…
so going super high capacity might not be worth that much… aside from saving space in the host and using less slots and thus giving the ability to expand with less infrastructure…
what’s best anyways… other than just an conceptual idea we have in our heads, but it never seems to be applicable to anything actually in reality.
each choice will have advantage and disadvantages, one should expect the iops to be the limitation, as it always is with hdd’s
on a late side note… do aim for getting 7200rpm drives… that makes a huge difference for iops
5400rpm or such drives aren’t that common in 3½" but they are most certainly there, and often with a decent price saving.
It doesn’t mean the Seagate and other manufacturers are really worse, the margin of failures is very low. But if you want take a look and compare the prices, I suggest you the WD Gold (for business) or Ultrastar (for datacenters). They are actually almost the same drives, but Ultrastar is often cheaper and more flexible, because you can format in 4k native if you wish, which could give slightly better performances in some cases (it can be formatted by HUGO program utility, you can ask to the support or find on Google, otherwise I can you give you the links for the latest version in any case). So, if you can afford, you can look for Ultrastar; https://www.westerndigital.com/products/data-center-drives/ultrastar-sata-series-hdd
The current last one is the Ultrastar DC HC550 (18tb), or the previous HC510-520-530 10, 12 and 14tb maybe cheaper today. HC550 comes with an innovation, the world first Energy-Assisted Magnetic Recording (EAMR) to improve the reading in the dense plates, but practically, it changes almost nothing in matter of performance but capacity. https://www.youtube.com/watch?v=lGvEV0qdnqk . The 20tb is not CMR but SMR, not convenient for us, so for desktops, 18tb is currently the max capacity available WD side. But, its a new drive, so of course, we don’t have enough feedbacks about the failing ratio, but only the previous generations as you can see on Backblaze. I personnally trust them. If you look for these drives, be sure to buy the right model for your device (for SATA drive, my part number was WUH721818ALE6L4, the characters are decrypted in the WD papers).
For me in europe, the best prices were usually in german shops. I use idealo.de to follow the prices.
I’m using WD Ultrastar drives. They seem good, though I did get three drives that were DOA, so…
in the past i’ve seen motherboards destroy hdd’s of certain models as fast as i could hook them up…
never seen it since tho…
i would recommend anyone not using 1st generation technology, it’s so rare i’ve ever seen anything shine that was brand new production, ofc once in a while that won’t be the case…
because the developers used the hell out of it before throwing it on the market.
and even if it works today… then how does it work when the warranty has run out.
but i will say i have been very happy with the 10 HGST (today WD Ultrastar)
duno what their AFR is but its like less than 1% i think…
Well, the backplane could have somehow damaged three of the 6 new drivs I bought, but that is also unlikely.
One drive was not recognized by Linux (it was recognized by MHDD, but any command resulted in an error), two others appeared to work fine, but failed extended SMART self tests (one drive then showed failure in the SMART attributes, the other appeared fine, but still had the log entry about a failed self-test). Warranty replaced all three drives.
the time it happened for me, it was a model or series issue with the sata controller… tho not enterprise gear… but still it happened…
was trying to setup raid, and every time i put in a drive of that certain series it would just instantly die, didn’t last literal minutes…
ended up replacing them on warranty and then running without using the onboard raid function in the controller… and then it seemed to run fine…
so yeah… one of those mysterious things that just happen from time to time with new tech…
I’m not sure this makes sense. Iowait is effectively idle time that is waiting for IO completion; it’s time the CPU could have been doing something but was waiting on IO instead. A faster CPU does not minimize iowait, in fact a faster CPU can increase iowait.
Faster storage (disks and/or controllers) decreases iowait.