BlueSCSI for SGIs?

[Jinroh]

I saw this reviewed earlier today for other PCs of the old era, and did not see a discussion yet.

It looks like it can be the cheaper solution than SCSI to SD, if you have the interest in checking it out.

Open Source, you can do by yourself, as a kit, or pre-assemble.

https://scsi.blue

 

[fraserN64]

It should work, the downside is I'm seeing reports of <=- 1 MB/s transfer rates.
For comparison it looks like an Indy with SCSI2SD v6 can achieve 4.5 MB/s.
And a spinning SCSI drive of 300 GB at 10K RPM can go as high as 25 MB/s
Reference: https://forums.sgi.sh/index.php?threads/scsi2sd-v6-performance-report.110/

 

[Elf]

I think speed is generally the caveat with the solutions that use a microcontroller (or full on CPU, like RaSCSI) to bit-bang SCSI using GPIO pins. While not necessarily technically elegant, they do result in a low cost solution in any case

 

[compgeke]

fwiw the bluescsi stock isn't a particularly ideal product. It's cost reduced to the point of kind of hampering performance and causing some weird compatibility issues - especially if it's not the only device on a chain.

A bit issue is there's no actual bus drivers on the board, instead running right off GPIO. This could explain some of the compatibility weirdness where it works on some machines and not others, or works on internal or not external and so on. Additionally, the STM32F103C8T6 can only go up to 25 mA per GPIO pin, and SCSI at its worst can do 24 mA for termination and similar tasks. The SCSI2SD guy is building a derivative product to address this issue by adding bus drivers and some other changes due to be released soon iirc.

The fact that they're almost all using microcontroller modules instead of soldering one to the board isn't ideal either. The SCSI2SD lost the DB25 unpopulated pads as they were acting as an antenna and causing SCSI issues. I can't imagine bus-driverless GPIO pins sticking up work any better than bare pads were.

The other thing is it's not meaningfully cheaper over alternatives like a SCSI2SD. A v5.2 (when they're in stock) is $62 assembled, while an assembled BlueSCSI will run you ~$50.

Now, the BlueSCSI does have some fun advantages with how it handles images. Being able to store disk images on an SD card vs partitioning a whole card is very useful. Especially if you want to emulate a CD drive (there's actually a project for this called Mac SD).

 

[Jinroh]

Very good responses. I did not see any spec from the source, so glad to hear some analysis.

 

[ankylo5aurus]

I didn't do any benchmarks so I have no hard data to offer, but I tried a BlueSCSI in an Indigo2 and it was perceptibly much slower than a 7200RPM UW SCSI hard drive (IBM ECE31708).

I do love the BlueSCSI and I'm fine with using it in a classic Mac that's going to be slow anyway, but it really holds back the SGI.

 

[Jinroh]

Thanks for the info. Great to have some first hand test.

 

[pixel]

@ankylo5aurus Was there any trick to getting the BlueSCSI working? I created an empty image with dd, and my Indy recognizes it, but when I try to partition it I get the error "sc0,1,0: cmd=0x1a Unexpected info phase 36, state 49. Resetting SCSI bus" repeatedly.

 

[ankylo5aurus]

@ankylo5aurus Was there any trick to getting the BlueSCSI working? I created an empty image with dd, and my Indy recognizes it, but when I try to partition it I get the error "sc0,1,0: cmd=0x1a Unexpected info phase 36, state 49. Resetting SCSI bus" repeatedly.

I recall getting some console spam like that but it worked, and I didn't do anything special. I'm out of town for a while so I can't take a look at it right now, but will try to remember to check when i get back.

 

[jgilje]

@ankylo5aurus Was there any trick to getting the BlueSCSI working? I created an empty image with dd, and my Indy recognizes it, but when I try to partition it I get the error "sc0,1,0: cmd=0x1a Unexpected info phase 36, state 49. Resetting SCSI bus" repeatedly.

Based on the "diskperf stats" thread you recently started, was there any trick to get past this error? I get a similar error on my Indy using a BlueSCSI.

 

[mashek]

I have one of these: https://store.rabbitholecomputing.com/ZuluSCSI-Compact-Homebrew-p/zuluscsi-compact-homebrew-kit.htm
I'll do diskperf stats soon.

 

[mashek]

Here's my hinv for my Indigo2:
CPU: MIPS R4400 Processor Chip Revision: 6.0
FPU: MIPS R4000 Floating Point Coprocessor Revision: 0.0
1 200 MHZ IP22 Processor
Main memory size: 192 Mbytes
Secondary unified instruction/data cache size: 1 Mbyte on Processor 0
Instruction cache size: 16 Kbytes
Data cache size: 16 Kbytes
Integral SCSI controller 0: Version WD33C93B, revision D
Disk drive: unit 1 on SCSI controller 0
Disk drive: unit 2 on SCSI controller 0
Disk drive: unit 4 on SCSI controller 0
Integral SCSI controller 1: Version WD33C93B, revision D
On-board serial ports: 2
On-board bi-directional parallel port
Graphics board: GU1-Extreme
Integral Ethernet: ec0, version 1
Iris Audio Processor: version A2 revision 1.1.0
EISA bus: adapter 0

IRIS 1# diskperf -W -n "ZuluSCSI Compact Homebrew" -c 250MB -t 60 /mnt4/test
#---------------------------------------------------------
# Disk Performance Test Results Generated By Diskperf V1.2
#
# Test name : ZuluSCSI Compact Homebrew
# Test date : Wed Nov 24 19:57:31 2004
# Test machine : IRIX IRIS 6.5 10070055 IP22
# Test type : XFS data subvolume
# Test path : /mnt4/test
# Request sizes : min=4096 max=4194304
# Parameters : direct=0 time=60 scale=1.000 delay=0.000
# XFS file size : 4194304 bytes
#---------------------------------------------------------
# req_size fwd_wt fwd_rd bwd_wt bwd_rd rnd_wt rnd_rd
# (bytes) (MB/s) (MB/s) (MB/s) (MB/s) (MB/s) (MB/s)
#---------------------------------------------------------
4096 5.96 4.47 0.74 26.04 22.44 25.90
8192 1.85 5.90 0.76 33.41 30.46 34.12
16384 1.93 6.45 0.93 39.89 38.10 41.81
32768 1.92 6.40 0.92 44.87 44.00 47.64
65536 6.42 4.46 26.91 47.36 47.64 51.36
131072 6.67 4.95 28.59 41.12 47.64 51.62
262144 6.20 5.03 26.44 38.55 44.64 47.69
524288 6.56 5.37 26.58 35.33 40.60 43.14
1048576 6.31 5.09 25.01 28.24 33.84 34.56
2097152 6.45 5.07 25.12 31.49 33.75 34.75
4194304 6.62 5.01 0.00 0.00 33.79 34.83
IRIS 4#

It's not the fastest thing and is slower to boot off of it than a hard drive
but it's quieter, and less taxing on older power supplies.

for comparison:
IRIS 1# diskperf -W -n "ST39173WC" -c 250MB -t 60 /mnt2/test
#---------------------------------------------------------
# Disk Performance Test Results Generated By Diskperf V1.2
#
# Test name : ST39173WC
# Test date : Fri Nov 26 01:45:59 2004
# Test machine : IRIX IRIS 6.5 10070055 IP22
# Test type : XFS data subvolume
# Test path : /mnt2/test
# Request sizes : min=4096 max=4194304
# Parameters : direct=0 time=60 scale=1.000 delay=0.000
# XFS file size : 4194304 bytes
#---------------------------------------------------------
# req_size fwd_wt fwd_rd bwd_wt bwd_rd rnd_wt rnd_rd
# (bytes) (MB/s) (MB/s) (MB/s) (MB/s) (MB/s) (MB/s)
#---------------------------------------------------------
4096 9.61 7.46 3.10 22.57 19.14 22.39
8192 3.42 7.90 4.71 30.38 27.00 31.19
16384 3.46 7.84 4.94 35.91 33.76 37.28
32768 3.46 7.73 2.06 19.88 35.57 43.37
65536 9.52 6.82 28.08 43.79 42.54 46.70
131072 9.87 7.27 26.32 38.90 42.56 46.95
262144 9.67 7.35 21.30 37.88 43.79 48.61
524288 9.08 7.78 23.89 39.26 39.29 42.55
1048576 9.64 8.23 24.76 28.02 32.74 34.76
2097152 9.77 7.32 23.22 30.64 33.18 34.95
4194304 9.64 7.32 0.00 0.00 33.22 34.86

IRIS 5# diskperf -W -n "HP ST3300007LC" -c 250MB -t 60 /mnt2/test
#---------------------------------------------------------
# Disk Performance Test Results Generated By Diskperf V1.2
#
# Test name : HP ST3300007LC
# Test date : Wed Nov 24 21:57:25 2004
# Test machine : IRIX IRIS 6.5 10070055 IP22
# Test type : XFS data subvolume
# Test path : /mnt2/test
# Request sizes : min=4096 max=4194304
# Parameters : direct=0 time=60 scale=1.000 delay=0.000
# XFS file size : 4194304 bytes
#---------------------------------------------------------
# req_size fwd_wt fwd_rd bwd_wt bwd_rd rnd_wt rnd_rd
# (bytes) (MB/s) (MB/s) (MB/s) (MB/s) (MB/s) (MB/s)
#---------------------------------------------------------
4096 8.29 6.09 3.99 27.58 23.72 28.24
8192 3.91 7.16 1.08 35.77 31.77 36.87
16384 3.85 7.31 5.41 40.90 38.23 42.88
32768 3.93 7.26 5.60 44.52 44.02 48.60
65536 8.47 6.33 27.70 44.92 45.59 49.09
131072 8.55 6.58 30.12 40.07 45.78 49.28
262144 8.05 6.51 24.14 36.35 44.03 46.83
524288 8.04 7.31 26.48 34.23 39.05 41.08
1048576 7.55 7.24 23.72 28.99 34.05 34.88
2097152 8.20 6.70 24.02 31.38 34.26 35.09
4194304 8.14 6.75 0.00 0.00 34.35 35.10

 

[mashek]

BTW, I bought several BlueSCSIv2 boards before realizing that they didn't work on my SGI's, at least for me in default configurations. If anybody has these working, please let me know.

 

[Mario9501]

BTW, I bought several BlueSCSIv2 boards before realizing that they didn't work on my SGI's, at least for me in default configurations. If anybody has these working, please let me know.

I got my BlueSCSI v2 to work on my IRIS Indigo and Indigo 2. I did have to use bluescsi.ini and change these settings:


[SCSI]
Quirks=0 ; Set quirks to Standard as Apple quirks are not needed.
EnableSCSI2=0 ; Usually have issues if this isn't disabled.
EnableUnitAttention=0 ; If you have an HDD image on the SD card, there's no reason to have this on.
Debug=1; This is usually unnecessary but could be helpful for issues.


I usually have the termination set to off on the BlueSCSI as well, and after that, I generally get an install done and am good to go.

 

[7spirals]

I've tested Acard SCSI emulators, SCSI2SD v5 and v6, ZuluSCSI, BlueSCSI, and PiSCSI. My speeds in general are: BlueSCSI = 1 - 2 MB/s, SCSI2SD v5 2-3MB/s, SCSI2SD v6 3-5 MB/s, ZuluSCSI = 7-9 MB/s, PiSCSI 3-9 MB/s, Acard 10 MB/s synchronous. The people at Acard won't even return emails about their SCSI emulators now, even though they still have them on the website in Taiwan. The ZuluSCSI may reach 10MB/s because they are using a real SCSI controller at least (along with an embedded RPI 2040 I think). The easiest to configure is the Acard (just pop in a SATA SSD). The second easiest is the SCSI2SD v6 (good GUI), ZuluSCSI is the most "unixy" and I like the way in which one interacts with it. The PiSCSI is fiddly and a bit buggy, but pretty fast all things considered. The BlueSCSI is probably the most rinky-dink and least configurable. It's like a cheapo version of the ZuluSCSI. If you are going to buy a device for emulating specific device strings and geometry, nothing beats the SCSI2SD v6 (which, of course, is now unobtainium).

I wish *sooooo much* that Acard still made their emulators. I have the 50-pin, 68-pin, and SCA versions and I bought them back-in-the-day for around $100 each. One is in my Tezro, one is in my Indy, and one is in R12k O2. I guess at least I got those, but damn, what about everyone else? What about my other Unix boxes? I don't think the SCSI emulation market is saturated by a damn sight. I wish I had been an electrical engineer and I could take it on as a project. As hot as I think I am with SCSI (I've written some drivers, done some CDB debugging, etc..) I'm still nowhere near able to create such a device. It makes me sad to see all the car part clones out there but nobody can make a damn SCSI emulator that's worth a damn? This is what's wrong with the world! hehe.

 

[legodude]

I agree with your points entirely. I use the PiSCSI and ZuluSCSI/BlueSCSI (and GoTek) extensively for my fleet of vintage machines. I don't mind the performance limitations of the ZuluSCSI on a Vaxstation or HP 712, but they are just too slow on an Octane or Ultra 60 so I am still running spinning disks. My few PCs all use IDE-SATA adapters with SSDs or IDE-SD adapters.

With the retro Mac/Amiga/synth scene, there is a decent market and interest in SCSI2SD-ish devices, but I just don't see the same level of need for an ACARD replacement device. The Macs needing UW SCSI level of disk performance had moved to IDE or have compatible SATA PCI cards, and I don't know that many synths use UW SCSI drives. That leaves mid to late 90s unix nerds as the main consumers and I just don't know that there are enough of us.

What I do wonder is what happened to all fo the ACARD devices? I don't know what the original use case was - were they ever shipped in a product like a disk array or workstation? What were production runs like?

mike

 

[7spirals]

I know for sure some of my own customers ordered dozens of them to install into MSA1000 units. They were all the SCA model. However, they are all still in service. I told the customer I would pay cash for them after they were retired. The problem is they don't have plans to do that this year. They have 128 of them (all purchased at $80 per unit back in the day, makes me want to cry). That customer doesn't have any of the 50-pin version. I have one in my R5k Indy. I wish I had 5 more of them for my other systems.

 

[legodude]

Wow, I didn't know that anyone still used FC SCSI disk arrays. I assumed that they would all be replaced at this point in time. I guess i need to start inspecting Ebay disk boxes to see if there are any ACARDs in the empty disk trays...

 

[7spirals]

I still have a lot of customers with both SCSI3 arrays and fiber channel. MSA1000's have good drivers and tools for OpenVMS and a lot of folks stick with them because of that.