Hi, just switched from a USB2 PC to a USB3 system and my 4TB drive is unable to sustain writes and even more writes and reads simultaneously.
I did not notice it until now because of the db locks. But i ran a benchmark with ubuntu DISKS tool and sometimes i get 6MB/s writes on the benchmark witch does simultaneously read/writes.
Tested sequential writes with FIO and they are also bad and unconsistents. I guess it is going to die soon.
It is a Toshiba Canvio Basic : stay away from this atrocity.
That said, i need another HDD 4TB at least. Do you have any models to advise ?
I prefer a 2.5" because it’s self powered and i run on a laptop so in case of power outage, the battery acts as a UPS for the whole setup
you will want large cache unless if you are going to use something custom on the computer… else stay away … far far away from SMR drives, then you want 7200rpm not 5400rpm which is often used for usb disks.
i guess the 2.5" idea works… but if the power is out… chances are other local network stuff is also down, ofc that depends on your connection, usually one gets more capacity in 3.5" drives for the same money.
Q1T1 are generally horrible on everything… even a good ssd won’t have more than maybe 40-50mb/s
maybe you can get 80 or so on the best of the best nvme ssd
the worst thing is it seems like storj is a lot like Q1T1 benchmarks… so disks really don’t have to do lots of things well… just that i think… but haven’t really gotten it tested out really well yet…
And this is a good benchmark, yesterday write were stucked to 5MB/s from 0 to 100%. My guess is the disk is going to fail. Every benchmark draws a new random pattern.
actually doesn’t look that bad imho, 6MB/s at high RW I/O isn’t bad… i’m running a zfs pool of 2 vdevs consisting of 5 drives in a raidz1 and a second set of 4 in raidz1 with a ssd for slog and a dedicated ssd with 600gb for L2ARC cache…
and even with 11 drives or 9 if we subtract the redundancy working in sync to supply one node with enough IO i cannot do anything on the pool without seeing the storagenode performance being affected…
so thinking you can make it run well on one drive is imo wishful thinking unless if you go all out ssd…
the problem is that its essentially small files of only 2mb spread all over the drive in the end… and you have to find and retrive them… something which for an idle disk takes about 6ms if its a 7200rpm drive…
what program is that?
been looking for a nice easy way to simulate a storagenode, i suppose that would do nicely
okay thanks ill go try and apt-get that, in the near future…
i can get to 85% successrate with my setup
yeah sometimes the older disks actually do IO better from what i’ve seen in the past… as capacity goes up but the technology often doesn’t change and thus the IO remains the name, but the area of the disk written or bits grow immensely and thus the IO becomes much less or something like that…
noticed it a while back when i was going through old harddrives and found that some of the oldest ones was actually the bests at dealing with high io… so ended up putting those in as operating system drives in some computer projects…
well from what i know you should be able to get those kinds of numbers doing true Q1T1 random RW
many factors to take into account …
if you scroll down to peak score you can see that even these ssd’s get 65mb/s and 40 something in mixed 4k read writes, and they are like magnitudes faster than hdds for this particular test
and this is essentially the kind of workload the storagenode is put under, but its a great site, you can look and see if you cannot find a hdd you find up to your specs… i can’t tell you exactly whats best its just an idea i got from the few months of testing i’ve done… so no real recommendation, i would recommend the site tho… lol
The thing is i don’t want to spend tons of money on this, i have a pretty solid FTTH internet though. Maybe i should switch my node to the 2TB drive and use the 4TB drive for cold storage, but it’s filled with 2.3TB node files, so how i can reduce it ?.. I noticed tons of deletion lasts days however.
the file system blocksize and sector sizes on the disk might also have a pretty large impact on performance with certain workloads.
some drives are mostly for storage… you might be able to find both drives or very similar on the site i linked, then you can compare their raw numbers…
No disqualification, but it can take a while. You’d also be limiting your potential income by limiting the size of the node. In my experience they eventually pay for themselves if you have a good connection. So that’s not the route I would go.
