Lenovo Thinkcentre/ThinkStation Tiny (Project TinyMiniMicro) Reference Thread

[adman_c]

Thanks a million @Parallax for making this awesome post.

I read through the thread searching for info on how people are managing to retain the use of a 2.5" hdd, while also using the PCIe card (as suggested by the note above) ... but I have not found this info.

Please, can someone point me to the specific post that covers this use case ?

Thanks in advance.

PS: I own an M920q

I'm currently booting Proxmox from a 256GB SATA SSD that has been "shucked" from its plastic casing and wrapped in electrical tape to prevent shorting. I have a 2x SFP+ 10GbE NIC in there as well, as shown in this album (actually a different NIC than is pictured here, but the idea is the same).

 

[gdaba]

Given that bottleneck, using a tiny as NAS is probably not the best ... still an incredibly useful platform for home labs, save on power costs (especially in these times) ...

Again, it's also worth considering whether this limitation exists only on some of the Lenovo tinies ... or all of them.

Thanks a million, @cadillac ... most useful insights you have shared ...

 

[gdaba]

I'm currently booting Proxmox from a 256GB SATA SSD that has been "shucked" from its plastic casing and wrapped in electrical tape to prevent shorting. I have a 2x SFP+ 10GbE NIC in there as well, as shown in this album (actually a different NIC than is pictured here, but the idea is the same).

interesting ... and I get the idea indeed. Thanks a million for sharing, @adman_c

 

[cadillac]

Given that bottleneck, using a tiny as NAS is probably not the best ... still an incredibly useful platform for home labs, save on power costs (especially in these times) ...

Again, it's also worth considering whether this limitation exists only on some of the Lenovo tinies ... or all of them.

Thanks a million, @cadillac ... most useful insights you have shared ...

Maybe i was a bit confusing. This is not tiny specific, but comes from the CPU - PCH topology used by Intel. Also, the DMI bandwidth is pretty high - around 3.3GB/s is usable from my benchmarks.The bottleneck affects the NVME raid1 the most, as when writing the mirror device both disks only have around 1.6GB/s usable, and that becomes the limiting factorfor the array.In an ideal situation the mirror's performance would be near the slowest performing device in it.

 

[cadillac]

Anyone here done a Windows install on M90q gen1? I tried both Windows 11 and Windows 10, neither of them recognize any disk devices and asks for drivers. Supplying the ones from here (Non-VMD recognizes the RST as compatible hardware) doesn't work either. Tried official RST 16-19 with no success. The SATA controller is in AHCI mode but tried in RAID and RST with Optane mode also.

 

[Helzy]

P90q gen1

Do you mean an M90q? How did you create your install media? Did you set BIOS defaults with F9 and save? Installing in Legacy mode might be causing issue. Try creating your install media in Rufus instead of the M$ media creator, for a UEFI supported install media with win10 or 11 .iso?

 

[cadillac]

Yes, corrected in the original post. Install media is created with Etcher and in the Boot selector it looks it is proper UEFI boot media, but i don't know if that can even be confirmed from inside the installer. I've tried Factory BIOS settings also. I will create Install media with Rufus now, and report back. Are You suggesting that this is a non-issue in Your setup?

 

[Helzy]

The only time I've seen what you describe is trying to perform an install in Legacy mode with win10 Rufus creates a secureboot compatible solution. (in this case win 11)

 

[cadillac]

Thanks for the info. Under Linux the instrumentation of the RST is done through EFI system variables. Most likely my install media is the culprit here... Will report back, and thanks again.

 

[Helzy]

ahh okay see your edit now, failing a BIOS default reset, sounds like either a bad drive or hardware issue. Only other thing that comes to mind is a drive that had some other file system or RAID setup previously, where the drive may have been cleaned with diskpart or a similar utility, but then never re-partitioned and formatted. Rufus can usually fix drives that diskpart cannot. If you can see the drive on another system in say a USB enclosure, then use Rufus to format it using the 'non-bootable' option.

 

[Helzy]

Hmm interesting, I was testing RAID setup recently on a p350 tiny and when looking for RST for Linux was informed.by the intel RST site that it's not supported in Linux.

 

[cadillac]

Hmm interesting, I was testing RAID setup recently on a p350 tiny and when looking for RST for Linux was informed.by the intel RST site that it's not supported in Linux.

Intel RST arrays (iMSM under the hood) are fully supported accessable by dm-raid and mdraid.Only the usage of NVME drives with RST is not supported by Linux , and since these tinies have only one SATA port (if they even have) that means RAID is not supported indeed. Currrently i'm very frustrated by this, and the only reason i'm fighting with MS install is to check how does NVME RST RAID performance compare to mdraid (if mitigates the DMI bottleneck when using two very fast drives). If it does offer significant performance upgrade i will look into possible solutions to enable RST Linux NVME support
And indeed the install media was causing the error with RST. Created one with Rufus and it is working good. Thanks for the help.

 

[Helzy]

I didn't get that far due to Intel fatigue (using ThinkStation Tiny's with 2x m.2 2280 support) as Intel deprecated RST support for non-intel branded drives on the P350 tiny, however, I'd love to know. I've seen anecdotal evidence that suggests write intensive RAID 1 mode does create a bottleneck, however in my use case DMI isn't a big concern. Look forward to your results.

 

[cadillac]

Unfortunately RST using Windows has no solution for the DMI bottleneck. The sequential write throughput of a RAID1 array is capped at 1500MB/s.

 

[Helzy]

Unfortunately RST using Windows has no solution for the DMI bottleneck. The sequential write throughput of a RAID1 array is capped at 1500MB/s.

So, can I ask what the experience of the bottleneck is? i.e., how does a write throughput like this affect desktop management in your use case? Does RAID 0 change the equation?

 

[cadillac]

I'm not sure i understand the question correctly. The actual use case where this becomes a problem for me is using it as a writeback cache in a SCSI SAN. Redundancy is a must (especially with flash storage) and 1.5GB/s throughput of the array becomes the limiting factor for the IO coming from IB/RDMA, where ingress bandwith is more than 4GB/s. The experience of bottleneck is that the DMI bandwith limits the througput of the array to 1.5GB/s, althrough the array could perform around 3.3GB/s if both members have enough BW.

 

[Helzy]

Ahh, okay. I thought perhaps you were using some sort of desktop management interface (DASH/vPro/AMT etc.,) and experiencing a bandwidth issue. I now have the picture in my head. I'd not actually considered using the Tiny's as iSCSI targets and use an older 3Par 7200 that I need to replace, ideally with a small footprint consumer NAS and 10GbE adapters with a mixture of Mikrotik routerboards and a NGFW. I love the idea of a small stack of TMM's and an ESXi vSAN, however, big fat NVMe drives are expensive and I'm more of a find used stuff and repurpose it guy. my Mikrotik switches are eBay finds as is my NGFW. I've only ever bought one new Tiny (the P350) and regret it to this day. the P340's are more versatile in my opinion, maybe not as fast but honestly my favorite find was a P920x. My small Tiny fleet does everything I need from home labbing to HTPC etc. the P350 is definitely a wicked performer but Intel made me frustrated.

 

[Helzy]

New Lenovo Nano on the way?

ThinkEdge SE30 Edge Client

When it comes to an IoT-ready platform, ThinkEdge SE30 is a game-changer. Powered by the latest Intel® technology and high-speed DDR4 memory, it breezes through the most demanding tasks with ease.

www.lenovo.com

 

[gdaba]

And indeed the install media was causing the error with RST. Created one with Rufus and it is working good. Thanks for the help.

So, it's most likely that your BIOS was set to force secure boot (this is now often the default on new systems) and your initial etcher written install media was blocked on account of that. Rufus defaults to writing secure boot capable boot media, but you could also have gone into BIOS and disabled secure boot.

 

[Helzy]

So, it's most likely that your BIOS was set to force secure boot (this is now often the default on new systems) and your initial etcher written install media was blocked on account of that. Rufus defaults to writing secure boot capable boot media, but you could also have gone into BIOS and disabled secure boot.

On tiny p350 my experience has been the windows installer (20H2) fails to recognize disks with secure boot disabled.