Fusion-io ioDrive II - 1.2TB+ drives , 0.09 or 0.08/GB

Dhiru

Member
Aug 14, 2016
45
11
8
Relisted item now has $7.90 shipping cost.

Does anyone know about the availability of driver for ESXi 6.7? This would make a good datastore for IOPS hungry VMs.
 

fossxplorer

Active Member
Mar 17, 2016
504
81
28
Oslo, Norway
This "westporter" seller is based on my experience known to try to get every extra $ out of whatæs being ofered. From what i read here, there is a push for price up both for BO and now with "shipping". :mad:
 

Northern

Member
Dec 2, 2015
87
36
18
56
This "westporter" seller is based on my experience known to try to get every extra $ out of whatæs being ofered. From what i read here, there is a push for price up both for BO and now with "shipping". :mad:
Uggg...!!! That's not what I wanted to here....
Well they are down to "only" 294 left... I guess they have sold about 15% of what they have made available...
Lets get serious about solidarity!! Only accept offers of $110 or less!! LOL
 

Oddworld

Member
Jan 16, 2018
64
29
18
121
My order of three arrived. Two needed firmware updates, but all drives appear to work fine.
 

fake-name

Active Member
Feb 28, 2017
167
125
43
71
Mine showed up. Both appear up-to-date wrt firmware (7.1.17, 116786).

Currently I'm trying to persuade one of the drivers to build on Ubuntu 18.04 (kernel 4.15.0-45-generic). You appear to not be able to view the drive write bits without the kernel module being installed.
 

acquacow

Well-Known Member
Feb 15, 2017
618
326
63
40
You appear to not be able to view the drive write bits without the kernel module being installed.
Yeah, that's because the device isn't a block device natively, it's a memory device.

You need the kernel module to provide the virtual block layer and do the address translation.

It's kinda the opposite of how virtual memory/swap work.
 
  • Like
Reactions: jfeldt

fake-name

Active Member
Feb 28, 2017
167
125
43
71
Yeah, that's because the device isn't a block device natively, it's a memory device.

You need the kernel module to provide the virtual block layer and do the address translation.

It's kinda the opposite of how virtual memory/swap work.
You can read a bunch of other stuff from the device without the kernel module (temperature, voltages, etc...), I would have initially guessed lifetime stuff would use the same interface.

Are write stats actually stored on on the flash? Doesn't that mean you could fiddle with the kernel module to trivially falsify all those numbers?


------

Got the driver chooching (using snuf/iomemory-vsl on ubuntu 18.04/Xen4.9.2).

Be aware you need the matching-version fio-xxx utilities. The updated library is based on 3.2.15, so if you install the 3.2.16 utilities, they fail to see the loaded driver, and are useless.

Code:
1 durr@xenbox:/media/Storage/IODrive/3.2.15$ sudo fio-status -fj | grep "total_physical_bytes_\|rated_writes_remaining_percent"
                    "rated_writes_remaining_percent" : "99.22",
                    "total_physical_bytes_read" : "142,530,735,115,624",
                    "total_physical_bytes_written" : "132,992,226,053,056",
                    "rated_writes_remaining_percent" : "99.22",
                    "total_physical_bytes_read" : "142,328,125,438,720",
                    "total_physical_bytes_written" : "132,771,327,338,800",
1/2 the total write endurance of the disk it's replacing (SSD 850 Pro 1TB) is 0.78% of the lifetime of these. Considering I've completely managed to use the lifetime writes of the 850 Pro in about 1 1/2 years, this should last me fine.

These things are hard core.
 
Last edited:
  • Like
Reactions: arglebargle

acquacow

Well-Known Member
Feb 15, 2017
618
326
63
40
You can read a bunch of other stuff from the device without the kernel module (temperature, voltages, etc...), I would have initially guessed lifetime stuff would use the same interface.

Are write stats actually stored on on the flash? Doesn't that mean you could fiddle with the kernel module to trivially falsify all those numbers?
The write stats are written somewhere in the FPGA. You'd need to hook up a JTAG reader/writer to really mess with any of that stuff. The fio-sure-erase utility literally wipes ALL of the flash, so that data isn't stored in there.

These things are hard core.
The ioDrives were really never intended to be stand-alone devices. Originally, they were going to be part of a flash array, but that device was going to be much more complex to produce. As a stop-gap measure, the flash modules were attached to pci-e carrier boards (they used to be two-piece PCBs) and then sold individually. We then discovered that things like mdraid worked just fine to build arrays out of the cards and that route was chosen for larger data requirements.

At the end of the day, the device is just an FPGA with some basic pci-e communication on it for low-level signaling, and then a bunch of fpga code flashed to it to turn the device into a drive, or memory extension, or whatever you coded your app to do with it (we had an SDK available). Most people just wanted to use it as a drive though.

The FPGA itself doesn't know how to do any of the drive primitives of read/write/seek, that's all in the driver level, which is why you can't use the device w/o any drivers installed.

The fio-utils do need to be matched version-wise with everything else in the Fusion-io world. They can get some low-level info from the FPGA, like the temp/device ID/etc. They will also let you do things, like run a command to wipe the entire drive, including the FPGA code, effectively bricking it...if you really wanted to.
 
Last edited:

fake-name

Active Member
Feb 28, 2017
167
125
43
71
The write stats are written somewhere in the FPGA.
FWIW, it's almost certainly on a EEPROM or similar on the card. The process nodes used to manufacture big FPGAs doesn't support producing EEPROM cells. My understanding is that Xilinx's "Non Volatile" FPGAs are actually manufactured by wirebonding a separate program flash IC to the actual FPGA die, and then the whole thing is encapsulated in a single package (There was a app note about how Xilinx changed the Spartan-3 AN flash IC die mid-production, it resulted in the need to update some files so they'd program correctly).

Dammit, now I want to take one of the cards apart to have a look-see.
 

acquacow

Well-Known Member
Feb 15, 2017
618
326
63
40
We also apparently only ever used about half of the FPGA, so it is theoretically possible to put other code in there and have the FPGA do work on the data on its way into or out of the card.

We had a few customers that wanted this as a feature, but it never happened. Lots of folks interested in doing data mining that way, or using the cards as network ring buffers for packet capture/etc.

If you knew a lot about FPGA programming, you could probably have some fun with the cards.
 
  • Like
Reactions: omgwtfbyobbq

fake-name

Active Member
Feb 28, 2017
167
125
43
71
If you knew a lot about FPGA programming, you could probably have some fun with the cards.
It'd be hard to get anywhere without a constraint file for pin definitions, at least.

I've been meaning to spend the time to learn to do linux kernel drivers for work-reasons, but I have too many projects already.
 

cw823

Active Member
Jan 14, 2014
331
116
43
Relisted item now has $7.90 shipping cost.

Does anyone know about the availability of driver for ESXi 6.7? This would make a good datastore for IOPS hungry VMs.
I installed the driver for 6.0 and all looks good in 6.7
 

Herou8

New Member
Mar 17, 2019
6
0
1
Hi,

Is this working in my desktop bootable storage?
I have windows 10 pro.
And my motherboard is Asrock Z77 Pro 4...

I planning to order one to me, because of good pricing and performance.
 

Herou8

New Member
Mar 17, 2019
6
0
1
IOdrive is Non booting pcie SSD drive.

Works fine in Windows 10 pro
So I have to have Sata ssd to booting and windows must be in there. Can I found that Fusion from my boot menu from bios?

Sorry, I am noob with ssds.