Indigo R4000 SCSI Performance - Printable Version
+- IRIX Network Forums (//forums.irixnet.org)
+-- Forum: SGI/MIPS (//forums.irixnet.org/forum-3.html)
+--- Forum: Hardware/Triage/Repair (//forums.irixnet.org/forum-11.html)
+--- Thread: Indigo R4000 SCSI Performance (/thread-4479.html)
+- IRIX Network Forums (//forums.irixnet.org)
+-- Forum: SGI/MIPS (//forums.irixnet.org/forum-3.html)
+--- Forum: Hardware/Triage/Repair (//forums.irixnet.org/forum-11.html)
+--- Thread: Indigo R4000 SCSI Performance (/thread-4479.html)
Indigo R4000 SCSI Performance - twix - 02-15-2025
I replaced a relatively modern SCSI disk in my Indigo R4000/100 with a ZuluSCSI device. This is is more much quiet solution, and hopefully puts a bit less stress on the ageing power supply.
Everything functions correctly, but performance is not stellar, with diskperf results not exceeding 3.5 Mb/s. I was expecting more. I tried to test the device in both sync and async mode (set with scsicontrol) but this does not seem to have any effect. The ZuluSCSI logfile indicates that the connection starts up in sync mode.
I then put the hardware SCSI disk back and tested that, and results were comparable to those of the ZuluSCSI device. Nothing higher than 3.5 Mb/sec. So it would seem that the type of SCSI device is not the issue.
Is this what I can realistically expect from my Indigo, or is something wrong here? Can anybody provide some comparison data for this machine?
RE: Indigo R4000 SCSI Performance - jpstewart - 02-16-2025
Keep in mind that the Indigo is using the oldest generation SCSI-1 interface with a maximum theoretical throughput of only 5MB/sec. So your performance figures are actually pretty good in my opinion.
RE: Indigo R4000 SCSI Performance - twix - 02-16-2025
I was under the impression that the WD33C93B controller is capable of up 10 mb/s. The WD33C93B manual states that there are two synchronous modes : 5 and 10 mb/s, selectable by setting a bit in the SYNCHRONOUS TRANSFER register. The ZuluSCSI log states that 10 mb/s synchronous transfers have been negiotiated.
RE: Indigo R4000 SCSI Performance - jpstewart - 02-17-2025
You're right about the 10 MB/sec rate. My mistake.
But I still don't think your numbers are that bad for the age of the machine.
RE: Indigo R4000 SCSI Performance - twix - 02-17-2025
The machine is perfectly usable, but I am curious why I am getting these numbers, and what the bottleneck is.
RE: Indigo R4000 SCSI Performance - weblacky - 02-17-2025
(02-17-2025, 09:59 PM)twix Wrote: The machine is perfectly usable, but I am curious why I am getting these numbers, and what the bottleneck is.
The bus speed number you're quoting is a theoretical maximum speed for the entire chain. Most of the time the devices don't go near that speed. For example it was pretty rare for a hard drive even in the last revision of parallel SCSI to actually be able to saturate the chain. I would think that back when this machine was made you could expect a hard drive to maybe go to maybe 5 MB/second. Remember we were still dealing with floppy disks back then.
First of all speeds depends on which ZuluSCSI you got, if it's NOT the red RP2040 version...well the lesser versions are going to be slower. But I would agree that I would expect closer to 5MB/s then you're getting
It also depends on the SD card and one thing that we haven't discussed and unfortunately I haven't had the set up or time to test would be alignment of the SD card and proper creation of the SD card's file system.
I don't know how many people in this forum remember when SSDs came out in the consumer market, but that's what we all had to worry about. We had to worry about NAND erase boundary byte alignment and file system cluster sizes if you wanted actual performance. One thing that was always recommended is that you download the SD card associations official formatter from their website. It's available for both windows and MacOS last time I looked.
The formatting application guarantees that they are at least placing the file system in the correct starting alignment to match the erase block boundary of the SD card. They also make best attempt to make sure the file system cluster size is an even multiple of the erase block.
I suspect that things like this as well as potential issues like lack of read ahead might be the reason that there are some performance misunderstandings with some of these new adapters. None of the adapters seem to have instructions that say you'll benefit from performing alignments. But I'd have to assume that they would. Especially because in my personal experience SD cards are always much slower than they're advertised for. So if you want to get whatever you're going to get out of them you definitely need to make sure that they're optimally formatted as they originally came out of the package. Normally SD cards are formatted optimally when they were packaged. But most people immediately blow that away not understanding that the manufacturer may have actually chosen a specific cluster size and a specific offset to gain you the most speed. That's why it's advantageous to take a look at these things when you get a new SD card before you start fiddling with it.
Either way I would start with the SD association's formatting application: https://www.sdcard.org/downloads/formatter/
From general memory I would say you should be using an exFAT with 128 KB cluster size, making sure you start on an block erase aligning boundary for your SD card (which the formatter should try do for you). However the secondary problem is you still need to make sure that your main XFS file system and swap in your disk image on the SD card is also properly aligned on the SD card.
It would definitely be helpful if somebody that knows how to read the FX output could chime in on how to check the offsets to make sure that they're starting at a flash erase boundary! Even with an aligned SD card exFAT formatting you could still have an unaligned image of the virtual hard drive that still trashes your performance. Technically even though you can't control how the disc label starts and that kind of stuff you can control the offset of the partitions/volumes. So I believe you can actually make a default disc label then erase some of the partitions then put them back in the correct order but manually specifying the starting offset for the critical partitions, such as swap and the main file system.
XFS wants to be a 4K cluster size anyway so that should still be fine if you're all aligned.
But I'm going let an expert tell me how I'm supposed to read FX layout output because I always get confused on what the numbers actually represent. But I am very interested to have the opportunity to try this because I firmly believe that part of the performance were not getting is not critically aligning the final file systems used under emulation in accordance with the underlying flash media, along with aligning the native file system on the flash media as well. This should only waste a tiny amount of space and considering that most people are buying 32 GB are higher for their SD cards I don't think you'll actually notice the loss in space versus the increase in performance you'll be gaining.
RE: Indigo R4000 SCSI Performance - robespierre - 02-18-2025
A hard disk's transfer speed is determined by track capacity * rotation speed. For example, a disk with 54 sectors per track has 27,648 formatted bytes per track. If that disk spins at 3600 rpm or 60 1/s, its transfer speed from the HDA to the disk's SCSI processor is (27648 * 60) = 1,658,880 bytes/sec. Sequential reads will approach that speed, minus overhead.
Another example: The Quantum Atlas XP 32150 is a 7200 rpm, 2.1 GB drive with between 80–134 sectors per track. Using the same formula, its internal transfer speed is between (80 * 512 * 120) = 4,915,200 bytes/sec, and (134 * 512 * 120) = 8,232,960 bytes/sec. The highest sectors/track are on the outer cylinders at the logical beginning of the disk, which is why swap partitions should be located there for best performance. This model disk was sold with fast-wide 20 MB/s SCSI interface.
RE: Indigo R4000 SCSI Performance - ebaigar - 02-18-2025
(02-15-2025, 02:00 PM)twix Wrote: Everything functions correctly, but performance is not stellar, with diskperf results not exceeding 3.5 Mb/s. I was expecting more.
Is this what I can realistically expect from my Indigo, or is something wrong here? Can anybody provide some comparison data for this machine?
As stated in other postings, the SCSI interface is probably the limiting factor. Have a look at this thread for some benchmarks ;-) Just ran the disperf on my Octane and got rates of 10MB/s and up to 30MB/s.
RE: Indigo R4000 SCSI Performance - twix - 02-18-2025
Thanks for all the answers and the info. Some points:
- I have used the proper SD formatter, using exFat on a Samsung 64Gb Pro Plus.
- my disk (Quantum Atlas V with converter to 50 pins) does not perform any better. The numbers are about the same.
- the diskperf results for the Indy in the thread referenced directly above are much better (>> 5 Mb/s for large blocks), and that uses the same SCSI controller.