SanDisk CloudSpeed Eco 1.92TB 2.5" SATA $120

[mrpasc]

It depends. If you continue to do more sequentiell writes, yes. If you start to hammer the drive with massive random write load it will not last that long.
The CloudSpeed Eco are „read optimised“ drives.
But the average homelaber should be good with them for some years. I think they will be die because of their capacitor’s drying out before the flash cells are gone.

 

[autoturk]

I think you can do that, but I don't think many has experience with drives that has reached 0% yet. I have one Intel s3500 with 30% life left and it still works great

It depends. If you continue to do more sequentiell writes, yes. If you start to hammer the drive with massive random write load it will not last that long.
The CloudSpeed Eco are „read optimised“ drives.
But the average homelaber should be good with them for some years. I think they will be die because of their capacitor’s drying out before the flash cells are gone.

thank you both for the perspectives!! While I have you both, I’m building a NAS with a 10G connection. I’m thinking of getting 8 of these and setting up truenas scale in a raidz1 with an optane slog. I should be able to saturate the link for reads and writes, right?

i imagine this is a much more economical option then trying to go with 4x enterprise NVMEs.

 

[mrpasc]

I’m thinking of getting 8 of these and setting up truenas scale in a raidz1 with an optane slog. I should be able to saturate the link for reads and writes, right?

For reads (if mostly from RAM ARC cache): yes
For writes: no. With RAID-Z1 you get barely the write rate of a single of your SATA ssd, around 500MB. And remember: a SLOG is NOT a write cache, it’s just a backup of the RAM based write cache for powerloss. So with sync writes disabled (like for a SMB share) it doesn’t help anything (will not be used), for sync writes like for a NFS share it helps to get near the write rate of the pool like it would be with async writes.
AGAIN: a SLOG is NOT a write cache (even it’s so named in official iX documents). Point.

 

[Markess]

thank you both for the perspectives!! While I have you both, I’m building a NAS with a 10G connection. I’m thinking of getting 8 of these and setting up truenas scale in a raidz1 with an optane slog. I should be able to saturate the link for reads and writes, right?

For writes: no.

Just curious, to advance the conversation, what's the best approach for @autoturk to boost write speeds? Is it the usual catch all performance improver for ZFS....more system RAM for a larger ZFS Cache?

 

[Shalita]

I ordered one of them 1.92tb for $46, it has 55k hours usage, but 99% life remaining.
Working perfect so far

 

[mrpasc]

Just curious, to advance the conversation, what's the best approach for @autoturk to boost write speeds? Is it the usual catch all performance improver for ZFS....more system RAM for a larger ZFS Cache?

Striped Mirror vdevs. Loose capacity but get higher write IOPs.

 

[GregH]

It depends. If you continue to do more sequentiell writes, yes. If you start to hammer the drive with massive random write load it will not last that long.
The CloudSpeed Eco are „read optimised“ drives.
But the average homelaber should be good with them for some years. I think they will be die because of their capacitor’s drying out before the flash cells are gone.

Pretty much my use case. It's in an enclosure replacing 2TB of USB 'portable' Toshiba spinning rust.
That said, by the time it was currency converted and shipping costs applied ~A$101,
I would have been better off sourcing one of these during Prime day last week

[Prime] Silicon Power P34A60 2TB PCIe Gen3 NVMe M.2 SSD $93.09 Delivered @ Silicon Power Amazon AU - OzBargain

C'est la vie

 

[SirCrest]

They now appear to be sold out. I never pulled the trigger. Maybe for the best as I can keep waiting until I'm ready to build.

 

[pixelBit]

They now appear to be sold out. I never pulled the trigger. Maybe for the best as I can keep waiting until I'm ready to build.

SANDISK CLOUDSPEED ECO SA620 1.92TB 2TB SATA III SSD SDLF1CRR-019T-1HA1 | eBay

<html> <head> </head> <body> </body> </html>

www.ebay.com

 

[smithse79]

Is it a terrible idea to get a group of these and a Netapp ds2246 to build a flash ZFS array?

 

[mrpasc]

I would double-check if the DS2246 can do Sata whiteout Sata interposer. Not really sure but if I remember correctly it can’t.
Hope others with more knowledge of NetApp shelves and sata compatibility chime in (myself only uses sas disks with them).

 

[smithse79]

Completely forgot about that. I have a 4246 that I believe has to have the interposers with SATA (I am also currently running SAS in that).

 

[UhClem]

I believe you can use SATA on, e.g., the 4246 (w/o interposers). But the shelf will be bandwidth-capped at 1700 MB/s, even with both controllers being fed. Might be that interposers will raise that SATA cap to 3400. Puzzling that the cap is 1700/3400, and not 2200/4400.

Maybe @BeTeP or @Stephan can provide some insight.

For this SSD, specifically, using a vanilla SATA3 connection, I've not been able to get spec'd performance

No matter what I tried, I couldn't exceed 21K read (4KB) IOPS;
and Read seq (128KB) @#jobs=8, qd=32 only got 465 MB/s.
(I didn't even mess with write's.)
Similar results on a 2nd drive.

An old Samsung 850 evo gets 85k @j4q8, and 520 M/s @j1q4

 

[BackupProphet]

That sounds weird, I get 85000 iops

sudo fio --filename=/dev/sdi --direct=1 --rw=randread --bs=4k --ioengine=libaio --iodepth=256 --runtime=120 --numjobs=4 --time_based --group_reporting --name=iops-test-job --eta-newline=1 --readonly
iops-test-job: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=256
...
fio-3.33
Starting 4 processes
Jobs: 4 (f=4): [r(4)][2.5%][r=314MiB/s][r=80.4k IOPS][eta 01m:57s]
Jobs: 4 (f=4): [r(4)][4.2%][r=320MiB/s][r=81.8k IOPS][eta 01m:55s]
Jobs: 4 (f=4): [r(4)][5.8%][r=320MiB/s][r=82.0k IOPS][eta 01m:53s]
Jobs: 4 (f=4): [r(4)][7.5%][r=320MiB/s][r=81.9k IOPS][eta 01m:51s]
Jobs: 4 (f=4): [r(4)][9.2%][r=320MiB/s][r=81.9k IOPS][eta 01m:49s]
^Cbs: 4 (f=4): [r(4)][10.8%][r=322MiB/s][r=82.4k IOPS][eta 01m:47s]
fio: terminating on signal 2

iops-test-job: (groupid=0, jobs=4): err= 0: pid=385030: Thu Jul 20 22:53:29 2023
  read: IOPS=81.5k, BW=318MiB/s (334MB/s)(4343MiB/13644msec)
    slat (nsec): min=1227, max=2019.7k, avg=47687.13, stdev=126115.15
    clat (usec): min=419, max=21590, avg=12505.40, stdev=1492.51
     lat (usec): min=671, max=21592, avg=12553.09, stdev=1499.67
    clat percentiles (usec):
     |  1.00th=[ 9372],  5.00th=[10159], 10.00th=[10683], 20.00th=[11207],
     | 30.00th=[11731], 40.00th=[12125], 50.00th=[12387], 60.00th=[12780],
     | 70.00th=[13173], 80.00th=[13698], 90.00th=[14484], 95.00th=[15008],
     | 99.00th=[16319], 99.50th=[16712], 99.90th=[17695], 99.95th=[17957],
     | 99.99th=[18744]
   bw (  KiB/s): min=297944, max=339128, per=100.00%, avg=326004.44, stdev=2026.27, samples=108
   iops        : min=74486, max=84782, avg=81501.11, stdev=506.57, samples=108
  lat (usec)   : 500=0.01%, 750=0.01%, 1000=0.01%
  lat (msec)   : 2=0.01%, 4=0.01%, 10=3.57%, 20=96.40%, 50=0.01%
  cpu          : usr=2.79%, sys=11.24%, ctx=760217, majf=0, minf=1064
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.1%
     issued rwts: total=1111754,0,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=256

Run status group 0 (all jobs):
   READ: bw=318MiB/s (334MB/s), 318MiB/s-318MiB/s (334MB/s-334MB/s), io=4343MiB (4554MB), run=13644-13644msec

Disk stats (read/write):
  sdi: ios=1109785/0, merge=0/0, ticks=821296/0, in_queue=821297, util=99.23%

 

[BackupProphet]

Ups, sorry that was an Intel S4500
Yeah I see that when I test against a raw device, I get 28k iops, however when I test 4kb random read on ext4 filesystem I get 76k iops

 

[Stephan]

@UhClem You tested what configuration exactly? I.e. which controller card on the server, which canister in shelf, which shelf, interposer yes/no?

And you say the Evo performed adequately in this configuration while the SanDisk didn't?

 

[UhClem]

@BackupProphet , thanks for checking. Odd that testing using a filesystem, vs. using the device directly (ie, /dev/sdX) actually improves performance, and so dramatically. But, seeing is believing (I get 74k w/j4q8).

@Stephan , my first paragraph was addressing the just-prior (diverged) discussion of the DS4246 regarding SATA usability, with vs without interposers. Then,

For this SSD, specifically, using a vanilla SATA3 connection, I've not been able to get spec'd performance ...

was meant to intro a back-on-topic separate issue (and maybe should have been a separate posting). I'll PM you tomorrow with the gory details on the "shelf issue".

 

[SirCrest]

SANDISK CLOUDSPEED ECO SA620 1.92TB 2TB SATA III SSD SDLF1CRR-019T-1HA1 | eBay

<html> <head> </head> <body> </body> </html>

www.ebay.com

huh... I literally searched directly for more of them and nothing came up.

But even still, I thought I was free of temptation.

 

[lhibou]

thank you both for the perspectives!! While I have you both, I’m building a NAS with a 10G connection. I’m thinking of getting 8 of these and setting up truenas scale in a raidz1 with an optane slog. I should be able to saturate the link for reads and writes, right?

i imagine this is a much more economical option then trying to go with 4x enterprise NVMEs.

As others have noted, mirrored vdevs in stripes are the way to go for speed. If you're planning to run eg: VMs on that pool that's what you'll wanna aim for.
For general file server duties where write speed isn't crucial, RaidZ is great.

 

[T_Minus]

If anyone has any of the ones with lower usage (<1PB and not insane hours) -- and wants to trade let me know how many you have and what parts you're interested in.... I have lots of DDR4 32GB RDIMM still to trade + lots of other goodies, so please message me.... ultimately looking for 30-50 of these more.