Just installed the Commell M2-225 2.5GbE adapter in my 5070s today, works great, and they're able to max out the entire 2.5g interface when vMotion-ing within ESXi.
(US) 90 dollar Wyse 5070 Thin client/mini-server?
- Thread starter WANg
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
I forget, how are people buying these?
I purchased them from Global American. Requested a quote and they got back to me pretty quick.
M2-225 - 2.5 Gigabit Ethernet LAN M.2 Module - Global American
The M2-225 is a low power, high performance M.2 2.5 gigabit Ethernet module. This is the perfect solution for small form factor or embedded PCs to expand functionality and speed.
globalamericaninc.com
Well... I don't know what I expected from this experiment but I put a Intel 900P 280gb in the PCI-E slot but it seems to be maxing out FIO 4k RANDOM READ / WRITE at 70K IOPS. I guess the CPU is too slow to keep the drive fed, I guess I'll swap it with a NIC and use this as an appliance. I really wish there was a Zen3 or 12th gen mini pc with a HHFL PCIE slot.
Pls post the (full)
fio command(s) you used, and their IOPS number(s).fio is a very well-engineered program. In fact, it is so flexible, with so many variables/options, that it is easy to (mistakenly) distort one's results.
In anticipation of setting up a 5070ext similar to what you describe (but with a P1600X), I am anxious, and motivated, to identify (and hopefully resolve) any (suspected) bottlenecks in its performance with Optane drives.
Hi, how much did they charge for 1? Thx MuchI purchased them from Global American. Requested a quote and they got back to me pretty quick.
M2-225 - 2.5 Gigabit Ethernet LAN M.2 Module - Global American
The M2-225 is a low power, high performance M.2 2.5 gigabit Ethernet module. This is the perfect solution for small form factor or embedded PCs to expand functionality and speed.
I figured that would only effect total bandwidth not latency for random 4K read and writes.
I'm not 100% what the FIO command PTS uses for their backed but here is what I have:
Sth Benchmarks [2307031-NE-STH96009388] - OpenBenchmarking.org
If you want to provide an FIO command to test I'd be happy to try it.
I'd expect at PCI-E 2.0 that we should be closer to 2.0 GB/S and real world around 1.7 GB/s.
Edit: looks like it's a CPU bound issue as adding more threads adds more performance until we run out of threads.
Bash:
nota@e5070-900p:/900p$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=4G --rw=randwrite --numjobs=4 --group_reporting
test: (g=0): rw=randwrite, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
...
fio-3.28
Starting 4 processes
Jobs: 4 (f=4): [w(4)][100.0%][w=583MiB/s][w=149k IOPS][eta 00m:00s]
test: (groupid=0, jobs=4): err= 0: pid=14574: Mon Jul 3 13:48:20 2023
write: IOPS=159k, BW=620MiB/s (650MB/s)(16.0GiB/26437msec); 0 zone resets
bw ( KiB/s): min=595048, max=732488, per=100.00%, avg=636345.54, stdev=14824.24, samples=208
iops : min=148762, max=183122, avg=159086.42, stdev=3706.08, samples=208
cpu : usr=15.79%, sys=56.46%, ctx=4163537, majf=0, minf=44
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=0.1%, >=64=0.0%
issued rwts: total=0,4194304,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=64
Run status group 0 (all jobs):
WRITE: bw=620MiB/s (650MB/s), 620MiB/s-620MiB/s (650MB/s-650MB/s), io=16.0GiB (17.2GB), run=26437-26437msec
Disk stats (read/write):
nvme0n1: ios=0/4192182, merge=0/5, ticks=0/75796, in_queue=75796, util=99.66%
nota@e5070-900p:/900p$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=4G --rw=randread --numjobs=4 --group_reporting
test: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
...
fio-3.28
Starting 4 processes
Jobs: 4 (f=4): [r(4)][100.0%][r=847MiB/s][r=217k IOPS][eta 00m:00s]
test: (groupid=0, jobs=4): err= 0: pid=14584: Mon Jul 3 13:48:45 2023
read: IOPS=233k, BW=909MiB/s (953MB/s)(16.0GiB/18023msec)
bw ( KiB/s): min=864768, max=1065488, per=100.00%, avg=933031.89, stdev=22370.48, samples=142
iops : min=216192, max=266372, avg=233257.97, stdev=5592.62, samples=142
cpu : usr=25.67%, sys=74.18%, ctx=3131, majf=0, minf=298
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=0.1%, >=64=0.0%
issued rwts: total=4194304,0,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=64
Run status group 0 (all jobs):
READ: bw=909MiB/s (953MB/s), 909MiB/s-909MiB/s (953MB/s-953MB/s), io=16.0GiB (17.2GB), run=18023-18023msec
Disk stats (read/write):
nvme0n1: ios=4169667/3, merge=0/1, ticks=60592/1, in_queue=60592, util=99.48%
Last edited:
My, my ... [very :-J (tongue-in-cheek)]
First, you belittle your cute little box, saying it is weak ... (publicly, no less)
[achieving 230k/160k r/w 4c64t/4c64t]
You really need to apologize to it; they're very sensitive, you know.
(it might run away from home ... or worse)
[/:-J]
330k/200k [2c3t/1c3t]
We'll see ...
First, you belittle your cute little box, saying it is weak ... (publicly, no less)
Then, you beat it, without mercy (all 4 cores, 64 threads each)
[achieving 230k/160k r/w 4c64t/4c64t]
You really need to apologize to it; they're very sensitive, you know.
(it might run away from home ... or worse)
[/:-J]
Thanks, but I plan to unbox mine in a few days. I expect to get about
330k/200k [2c3t/1c3t]
We'll see ...
From my recent testing the best I've been able to get is still that 233k/159k read/write bandwidth, I think the problem is that the CPU is bottleneck the benchmark, you get better performance for the first few seconds when it's boosting. The performance tops out at 4 jobs and 3+ iodepth, anything beyond that doesn't improve performance.
I think the step forward will be to figure out how to get the CPU `intel_pstate` into active mode so I can set `performance` governor and maybe eek out a bit more performance.
Wondering if anyone has tried using a H1110 (LSI SAS9211-4I IT Mode, LSI SAS2004 controller) on the 5070 (extended) to connect additional drives. The drives will be powered (and mounted) externally but I'm trying to see what I may be missing from a compatibility perspective if I go that route... Objective is to set it up as a NAS under Proxmox...
From a digital/electronic perspective, it should be A-OK.
But, analog/mechanically, it depends ...
How will you get a SAS2 signal cable out of the box? (whether miniSAS-to-miniSAS or mini-SAS-to-4x breakout)?
Will you commit the Option-port (the punchout below the PCIe slot bracket opening)?
And, even if so, CAN you get there? (from the sff-8087 connector location on the card)
====
"Measure twice, cut/buy once."
mini-SAS to 4x SATA.
If I keep the wyse & drives in separate cases, I was actually thinking of removing the Parallel port and using that opening/exit instead so the cable don't need a "sharp" turn under the card to reach the option port which may make the run slightly longer.
Good reminder
I saw that there are either 3.3ft/1m or 1.6ft/0.5m breakout cables so considering the card is 3.12in long, I may be able to use 5in internally and have around 34in left to enter the other case...If I end up moving everything to a single case, then it should work as well, but I definitely need to "practice" with a dry run
Do you say that mostly because of the cable routing or something else? Do you know of a specific 4-external card that is PCIe x4?
Thanks!
Another option is you can remove the Dell logo, which will leave a round hole in the front.
4I = internal, so any model with 4e or 8e is external
Hold that thought for a minute.
Yes, much better (but moot).
Gack! That dichotomy of aesthetic is painful.
Separate is the way to go, and not just for form, but also for function.
[my prior post was actually a "set-up" for this ...]
@Fritz was on the right track, but ...
The clear winner is the Dell H200e, a LSI 9200-8e equiv[**]; uses the SAS2008. E.g., [Link].
It uses just 1-2w more than the -4i, but easy growth to 8 drives. Now, here's why external is a win: Instead of some "ghetto claptrap duopoly", you have a nice tidy little cottage with an RV parked nearby. And, you can just detach the RV, easily--maybe to mow the lawn--or park it next to a different host. Just a "safe remove"; life in the cottage doesn't even notice. And re-attach is even easier.
[**] Fret not the x8 card; the WyseExt slot is open-ended
.
WOW! oh my... I've been looking at the board and checking routing options and... completely missed that
thanks a lot! this opens a lot of possibilities now
That makes a lot of sense - thank you for the insight!
I'm glad I've posted here before buying all pieces - back to the drawing board!
To which, I had replied:
I made this estimate based on tests in my workstation (HP Z4G4 w/W-2145 CPU**) with the SSD's PCIe slot (bios-)limited to Gen2; and factoring those results relative to the Passmark (single-thread) #s (W2145:J5005) of 2609:1205. [Yes, Passmark isn't gospel; but "close enuf fer guvmnt work"?]
Well, my real #s (on 5070ext) are 278k/121k
(numjobs=4,iodepth=3/nj=3,iod=3)Read script,
onetst.sh:
Code:
#!/bin/bash
[ $# -ne 3 ] && echo Usage $0 numjobs iodepth BLOCKSIZE && exit 1
fio --name=onetst \
--filename=Test \
--filesize=20g --rw=randread --bs=$3 --direct=1 --overwrite=0 \
--numjobs=$1 --iodepth=$2 --time_based=1 --runtime=10 \
--ioengine=libaio --gtod_reduce=1 --group_reportingonetstw.sh:
Code:
#!/bin/bash
[ $# -ne 3 ] && echo Usage $0 numjobs iodepth BLOCKSIZE && exit 1
fio --name=onetstw \
--filename=Test \
--filesize=20g --rw=randwrite --bs=$3 --direct=1 --overwrite=0 \
--numjobs=$1 --iodepth=$2 --time_based=1 --runtime=10 \
--ioengine=libaio --gtod_reduce=1 --group_reporting
Code:
@tiny:/mnk [ 1284 ] # onetst.sh 4 3 4k
onetst: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=3
...
fio-3.30
Starting 4 processes
Jobs: 4 (f=4): [r(4)][100.0%][r=1088MiB/s][r=279k IOPS][eta 00m:00s]
onetst: (groupid=0, jobs=4): err= 0: pid=9918: Mon Jul 17 13:59:06 2023
read: IOPS=278k, BW=1087MiB/s (1140MB/s)(10.6GiB/10001msec)
bw ( MiB/s): min= 1086, max= 1090, per=100.00%, avg=1088.22, stdev= 0.45, samples=76
iops : min=278072, max=279150, avg=278584.74, stdev=115.00, samples=76
cpu : usr=31.44%, sys=68.18%, ctx=18199, majf=0, minf=50
IO depths : 1=0.1%, 2=100.0%, 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=2782328,0,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=3
Run status group 0 (all jobs):
READ: bw=1087MiB/s (1140MB/s), 1087MiB/s-1087MiB/s (1140MB/s-1140MB/s), io=10.6GiB (11.4GB), run=10001-10001msec
Disk stats (read/write):
nvme0n1: ios=2754733/0, merge=0/0, ticks=41651/0, in_queue=41651, util=99.01%
Code:
@tiny:/mnk [ 1326 ] # onetstw.sh 3 3 4k
onetstw: (g=0): rw=randwrite, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=3
...
fio-3.30
Starting 3 processes
Jobs: 3 (f=3): [w(3)][100.0%][w=475MiB/s][w=122k IOPS][eta 00m:00s]
onetstw: (groupid=0, jobs=3): err= 0: pid=10311: Mon Jul 17 15:06:53 2023
write: IOPS=121k, BW=474MiB/s (498MB/s)(4746MiB/10002msec); 0 zone resets
bw ( KiB/s): min=481344, max=489144, per=100.00%, avg=486326.32, stdev=717.49, samples=57
iops : min=120336, max=122286, avg=121581.58, stdev=179.37, samples=57
cpu : usr=13.81%, sys=61.09%, ctx=1286106, majf=0, minf=26
IO depths : 1=0.1%, 2=100.0%, 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=0,1214906,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=3
Run status group 0 (all jobs):
WRITE: bw=474MiB/s (498MB/s), 474MiB/s-474MiB/s (498MB/s-498MB/s), io=4746MiB (4976MB), run=10002-10002msec
Disk stats (read/write):
nvme0n1: ios=0/1202028, merge=0/0, ticks=0/19113, in_queue=19113, util=99.01%** very similar to YOUR Dell_5820 w/W-2140b. You could test the 900P there. bios-limiting the slot to Gen2... but, why?
I have a dual E5-2670v4 blade with a U.2 900P drive in it and I'm able to push 596K/454k with 4 jobs 3 iodepth and 4k blocks. So we are no where near the drive limit lol.** very similar to YOUR Dell_5820 w/W-2140b. You could test the 900P there. bios-limiting the slot to Gen2... but, why?
I think the PCI-E 2.0 is a J5005 SOC limitation.
I think this this box is just going to get a quad gig NIC and call it a day. As for my pile of Intel 900P AICs that I bought because I thought they were U.2... maybe there is some cursed way to use USB 3.2Gen2 NVME adapters using a m.2 to pcie 4x riser to get the 900p working as an external drive.