What Should be the Next Stress Test's Goal?

[wuyanren]

Stable program, is a good program

[clayop]

Based on the current transaction capability cap as BM mentioned, I am not able to run 300 TPS / 30 minutes stress test.

Instead, I would like to run a 100 TPS / 1 hour stress test, as a final stress test before the 2.0 launch.

It may requires 15~25 VPS (I expect 4~7 TPS per node) and about 400k CORE.

I will determined the test schedule with the cooperation with @bytemaster or other devs and announce it here - https://bitsharestalk.org/index.php/topic,18684.0.html.

[bytemaster]

For BTS the only thing we  need for the next 6 months is 100 TPS and a stable / reliable set of witnesses. 

For our test networks we can continue to push the limits, but that will have to be done with high quality computers connected to high-bandwidth internet, potentially with very low latency nodes.

At 1000 TPS the bandwidth requirement of the nodes is already near 2 MB per second (assuming the producer broadcasts it to 8 peers and each block is 333KB).    Even at 1MB / sec bandwidth, the minimum latency for a block to TRANSFER from the producer to the first peer is 0.5 seconds.   If the producer is dividing their bandwidth 8 ways then it will require either 4 seconds or 8MB /sec bandwidth just for the blocks.    The actual bandwidth requirements are double that because transactions are sent separately.

In other words, until we rework the protocol to avoid rebroadcasting transactions when we broadcast blocks any peer that is unable to sustain those kinds of data rates will be a problem.

Also, the networking code currently uses more CPU than the blockchain validating code.   Fortunately it runs in its own thread.   Those running on a single core VPS with 1 GB of ram will not be able to keep up.   Those running on a home PC will also not be able to keep up.   

[puppies]

If you're doing a stress test, it would be a lot more useful if you list the hardware in the VPS units used, then what the bottleneck is that you saw...

Yeah.  I was hoping to discuss it in the other thread.  Thats why I didn't address it.  I am of the opinion that the bottleneck is single core processor speed.  I think an i5 at home or a $40 a month vps would be a good starting point.  Ultimately we just need to keep performance above 66%.  Learning which nodes are unable to keep up will be just as valuable as learning which nodes are able to keep up.

[r0ach]

If you're doing a stress test, it would be a lot more useful if you list the hardware in the VPS units used, then what the bottleneck is that you saw...

[puppies]

I say push it harder.  Maybe we try a run with just higher powered servers, and no fees.  I want 20k tps.

That can be another private testnet, imo.

Great idea
https://bitsharestalk.org/index.php/topic,18743.new.html#new

[clayop]

I say push it harder.  Maybe we try a run with just higher powered servers, and no fees.  I want 20k tps.

That can be another private testnet, imo.

[puppies]

I say push it harder.  Maybe we try a run with just higher powered servers, and no fees.  I want 20k tps. 

[wuyanren]

I think it is now mainly stable, within six months, 200tps can be

[cube]

No. Do not scale back.  But wait for BM and team to restore the testnet and fix it up for even higher TPS.  I think it is pre-mature to think of the next TPS goal when the dev team is away for the weekend.  Let's patiently wait for BM.

A note on the load test preparation progress, I am developing test scripts for a load test that includes other functions such as 'create user accounts', 'create asset'.  I hope this would provide a more comprehensive appraoch to the current stress test based solely on a fixed accounts  'core transfer' operation.  I will make the announcement soon.

[tbone]

Scale back and focus on stability


Sent from my iPhone using Tapatalk

I agree with this.  Also, perhaps going forward all spamming should be coordinated with bytemaster so he can a) observe the network as the spamming occurs, and perhaps more importantly b) so he can restart the testnet if it goes down.  This way we can avoid losing more precious time with the network being down. 

[clayop]

wait the stress test is happening? is there a place I was supposed to send the 2500 bts I offered to donate to the cause?

It's not for 100k tps.

[unreadPostsSinceLastVisit]

wait the stress test is happening? is there a place I was supposed to send the 2500 bts I offered to donate to the cause?

[clayop]

Scale back and focus on stability

Good point. We should balance in between

[sittingduck]

Scale back and focus on stability


Sent from my iPhone using Tapatalk

[clayop]

The result proved that 1000 TPS for 15 min was too harsh for current witness servers (but some high-end witness VPS survived)
In the next testnet, what is your preference for target TPS and its duration?