Solidigm Linux support

taggart · May 4, 2023, 6:54am

The new video about Solidigm was released today. Very interesting… I want to know how well they work in Linux. Here is what I found when I searched:

Phoronix P41 Plus linux performance testing (Sept 2022)
Phoronix P44 Pro linux performance testing (Dec 2022)
The Solidigm Storage Tool is supported under Linux, latest is 1.7 March 2023
The Arch wiki page has some good info on Intel/Solidigm
The nvme-cli tool has a solidigm plugin which seems to be mostly about SMART stats and product specific logs. See commits. They also has commands and documentation specific to Intel, some of which might actually be for this class of drives.
fwupd seems to have some support for Solidigm. Maybe due to Dell and maybe just for Intel 670P and p421plus?

Here are the questions I still have:

Do these devices have Power Loss Protection?
How well do they work when using LUKS or other encryption types? I remember some very early SSD controllers (sandforce maybe?) did some tricks to improve performance and reduce wear, like block deduplication, Windows specific things, etc (this was all pre-TRIM I think) but those fell down in the face of writing essentially random data. Still, the firmware could probably still optimize around which blocks seem to get used the most…
Is there any plan for an upstreamed Linux driver?

Utini · May 6, 2023, 3:14pm

I am using the P44 Pro 2TB on Arch with my T14s Gen3 AMD.

So far everything works perfectly fine but I wonder if the “special windows solidigm driver” is available in Linux as well?

wwed26 · May 6, 2023, 10:01pm

@taggart Thank you you kind soul I have four of these bad boys and was going to make a few posts on them.

My write speed is poopy, now I have something to start chewing on.

0xDE57 · May 7, 2023, 2:30am

I am wondering this too about the driver. Just picked one up for the new build

Utini · May 7, 2023, 7:52am

Hm I cant really see a speed difference between Arch vs Win10. You guys got benchmarks that show differences?

EniGmA1987 · May 7, 2023, 6:16pm

One of the things that was mentioned on the L1 show was that you may not necessarily see a performance gain you are getting in a typical benchmark. The special Windows driver is supposed to be able to prefetch additional data during something like a game load where the game keeps requesting a bunch of tiny data bits, but the addresses are all sequential, the driver will start grabbing a larger sequential block to improve performance and keep the extra data cached to give it to the game as those next addresses keep coming up. A benchmark wont show that because sequential load is already sequential and random load is fully random and not a bunch of “randoms” at sequential addresses. It would be interesting to see this added to a benchmark like Crystal Disk to simulate what some drives can do with their software side vs others that dont have a software component.

wwed26 · May 17, 2023, 12:21am

Which benchmark are you using. I get really crappy sustained RW but researching if it’s for a separate reason.

Utini · May 19, 2023, 4:16am

KDEDiskMark and CrystalDiskMark

wwed26 · October 31, 2023, 9:01pm

Has anyone had luck fine tuning their drive in Linux. Looking for tips given my performance is subpar

0xDE57 · November 3, 2023, 6:17am

what sorts of numbers should we be expecting?

KDiskMark is giving me some trouble:
‘Benchmark could not be prepared. Bad FIO output’
Screenshot_20231102_235207

got it to go:
Screenshot_20231102_235047

sst output:
sst show -ssd

-sensor for PowerOnHours or Temperature.
sst show -ssd 0 -sensor
Screenshot_20231103_000301

alkafrazin · November 3, 2023, 6:39am

I got FIO errors on some Linux 6.5 versions, and had to skip those. Check if you’re on a revision of 6.5.3 or later, and see if there’s an update. I think 6.5.8 it was resolved by then.

0xDE57 · November 3, 2023, 6:59am

thanks. updated kernel. No longer gets FIO error, but overall slightly worse numbers on this run. same default settings.

Screenshot_20231103_005654

and KDiskMark still seems to be struggling to finish runs

alkafrazin · November 3, 2023, 11:30am

check sudo dmesg, FIO may still be crashing out and just not pushing the error.

Exard3k · November 3, 2023, 12:39pm

Use FIO in the command line. If there is an error, FIO tells you what it is all about. KDiskmark is a nice GUI tool if you’re lazy with typing and looks very similar to CrystalDiskmark. I still miss some flags I occasionally use.

86-92 MB/s on 4k randwrite QD=1…that’s really good

0xDE57 · November 6, 2023, 12:23am

correct you are. dmesg says:

[75580.222231] fio[162402]: segfault at 0 ip 0000000000000000 sp 00007ffcb7309d48 error 14 in fio[563f98278000+1b000] likely on CPU 16 (core 4, socket 0)
[75580.222239] Code: Unable to access opcode bytes at 0xffffffffffffffd6.
[75602.197847] kdiskmark_helpe (162466): drop_caches: 1
[75602.755596] kdiskmark_helpe (162466): drop_caches: 1
[75603.317846] kdiskmark_helpe (162466): drop_caches: 1
[75603.872592] kdiskmark_helpe (162466): drop_caches: 1
[75604.438325] kdiskmark_helpe (162466): drop_caches: 1
[75604.464101] fio[162606]: segfault at 0 ip 0000000000000000 sp 00007fff68f75e78 error 14 in fio[55672c6da000+1b000] likely on CPU 17 (core 5, socket 0)
[75604.464109] Code: Unable to access opcode bytes at 0xffffffffffffffd6.

0xDE57 · November 6, 2023, 12:30am

What are some sensible defaults to test? I just ran this based on a search:

fio --name=random-write --ioengine=posixaio --rw=randwrite --bs=4k --numjobs=1 --size=4g --iodepth=1 --runtime=60 --time_based --end_fsync=1

Summary

random-write: (g=0): rw=randwrite, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=posixaio, iodepth=1
fio-3.35
Starting 1 process
random-write: Laying out IO file (1 file / 4096MiB)
Jobs: 1 (f=1): [w(1)][100.0%][w=270MiB/s][w=69.1k IOPS][eta 00m:00s]
random-write: (groupid=0, jobs=1): err= 0: pid=169102: Sun Nov  5 17:23:34 2023
  write: IOPS=69.0k, BW=270MiB/s (283MB/s)(15.9GiB/60491msec); 0 zone resets
    slat (nsec): min=411, max=95480, avg=941.39, stdev=349.24
    clat (nsec): min=271, max=7285.5k, avg=12151.65, stdev=8615.97
     lat (usec): min=9, max=7286, avg=13.09, stdev= 8.67
    clat percentiles (nsec):
     |  1.00th=[ 9792],  5.00th=[10048], 10.00th=[10432], 20.00th=[10688],
     | 30.00th=[10944], 40.00th=[11456], 50.00th=[11840], 60.00th=[12480],
     | 70.00th=[12864], 80.00th=[13376], 90.00th=[14272], 95.00th=[14912],
     | 99.00th=[16768], 99.50th=[18048], 99.90th=[28544], 99.95th=[33536],
     | 99.99th=[45312]
   bw (  KiB/s): min=23200, max=357624, per=100.00%, avg=290699.16, stdev=51811.67, samples=114
   iops        : min= 5800, max=89406, avg=72674.77, stdev=12952.91, samples=114
  lat (nsec)   : 500=0.01%
  lat (usec)   : 10=4.34%, 20=95.38%, 50=0.27%, 100=0.01%, 250=0.01%
  lat (usec)   : 500=0.01%, 750=0.01%
  lat (msec)   : 4=0.01%, 10=0.01%
  cpu          : usr=10.09%, sys=25.97%, ctx=4175471, majf=0, minf=20
  IO depths    : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
     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=0.0%
     issued rwts: total=0,4175222,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=1

Run status group 0 (all jobs):
  WRITE: bw=270MiB/s (283MB/s), 270MiB/s-270MiB/s (283MB/s-283MB/s), io=15.9GiB (17.1GB), run=60491-60491msec

Disk stats (read/write):
  nvme0n1: ios=59/1827256, merge=0/5068, ticks=92/215704, in_queue=215873, util=15.61%

I’m not sure what exactly I am looking at, but anecdotally I am happy with the overall performance of my machine so far during regular use.

Will run any benchmarks and post results for comparison.

wwed26 · November 14, 2023, 9:58pm

Getting crummy numbers on bench marking. I don’t know where to start. Can anyone shed some light on this.