slowish neek software download speed

I tired an EtherBlaster in place of a USB blaster hoping that it would download software to a Nios2 a bit faster, but there is no significant improvement. Maybe 13 kBps (104 kbps) could be a bit slow for the portential of an 8 MHz jtag connection, but perhaps there isnt any way to speed up the edit-compile-download cycle other than to implement an Ethernet boot.

Loading section .text.entry, size 0x20 lma 0x4000

Loading section .text.except, size 0xf0 lma 0x4040

Loading section .text.tc, size 0xb10 lma 0x4130

Loading section .text, size 0x54fc4 lma 0x10000000

Loading section .rodata, size 0x7b98 lma 0x10054fc4

Loading section .rwdata, size 0x1ae8 lma 0x1005cb60

Start address 0x4000, load size 389732

Transfer rate: 13 KB/sec, 397 bytes/write.

I had another idea to speed this up, but perhaps there are other limiting factors.

Ref: http://ecos.sourceware.org/fom-serv/ecos/cache/87.html

> set remote memory-write-packet-size 1000

> show remote memory-write-packet-size

The memory-write-packet-size is 1000. Packets are limited to 424 bytes.

> set remote memory-write-packet-size fixed

> show remote memory-write-packet-size

The memory-write-packet-size is 1000. Packets are fixed at 1000 bytes.

> load

Loading section .text.entry, size 0x20 lma 0x1000

Loading section .text.except, size 0xf0 lma 0x1040

Loading section .text, size 0x7d834 lma 0x8000000

Loading section .rodata, size 0x7ef0 lma 0x807d834

Loading section .rwdata, size 0x211c lma 0x8085740

Start address 0x1000, load size 555344

Transfer rate: 13 KB/sec, 969 bytes/write.

I see now that the USB blaster appears to be much, almost an order of magnitude, faster than the Ethernet blaster, or the direct USB connection to the NEEK board. This was against a different board than the NEEK, but the above posted delays for the Ethernet blaster are against the same board as the results below. Thats odd. Why would the Ethernet blaster be ten times slower than the USB blaster. This is odd considering that the speeds observed are much lower than full throughput on either link type (so presumably the Ethernet blaster is the limiting factor).

Loading section .text.entry, size 0x20 lma 0x1000

Loading section .text.except, size 0xf0 lma 0x1040

Loading section .text, size 0x7dce4 lma 0x8000000

Loading section .rodata, size 0x7f34 lma 0x807dce4

Loading section .rwdata, size 0x20dc lma 0x8085c20

Start address 0x1000, load size 556548

Transfer rate: 111 KB/sec, 398 bytes/write.

These tests were run on Ubuntu Linux.

I tested the USB-Blaster vs the Ethernet-Blaster and found the Ethernet device to be only slightly faster, but it was faster. It may have single stepped a little snappier too. It was not enough to switch because for me using Ethernet requires running Quartus to get the debug server running. I could never load Eclipse and debug without running Quartus first.

Also "real" Altera USB-Blasters are *way* faster than clone (e.g. Terasic) Blasters. Another case of you get what you pay for.

Bill A