Core_21 traffic query

[billford]

Absolutely nothing to do with the current instability problems

It occurred to me that some of these WUs (eg P9625-43) are pretty big, up around 50MB, and it wouldn't surprise me if bigger ones are on the way.

A reasonably quick GPU can process one of these in around 3 hours or less, in the worst case that's about 12GB download per GPU per month... even with three rigs that wouldn't be a problem for me but could be for some with limited bandwidth from their ISP.

Possibly a case for resurrecting the max-packet-size flag to it's (I assume) original function? I appreciate that this might (depending on flag/card type/client version interactions) currently clash with some Core_15 users, so maybe for when Core_15 goes EOL?

[toTOW]

I agree with your idea of resurrecting "max-packet-size" setting.

I'm also thinking about people with unlimited but slow connections, that might not be able to complete the download/upload of the WU before the server timeout occurs (15 minutes typical, 30 on servers with big DL/UL WUs).

[Nathan_P]

Good idea, especially when people are stuffing 2-3 gpu's in a rig, even a generous data allowance would soon get chewed up.

[ChristianVirtual]

Similar to here: https://www.reddit.com/r/foldingathome/ ... partially/
Though not data-driven but restricting on runtime ... Competing for a parameter

[billford]

Competing for a parameter

That was my thought- same request, different motives

[toTOW]

max-packet-size has always been set up based on data transfer sizes ... so it's not competing, it's the historical meaning of the parameter

[MARSTG]

I have a question on the same subject : how much is in general the size of a core 18 or 21 WU? I think I have seen in a log something like 14.5 MB for the download of a Core 18 WU but how much is the upload? I am not on a slow connection but on one tethering from cellphone and I would like to get an informed answer to see how much I can fold.

[bruce]

The concept of max-packet-size originated when uploads and downloads were limited by 5kb modems. Folding has generally completed a huge number of studies which could reasonably be managed by the commonly used settings of SMALL/NORMAL/BIG. Most current research studies now involve much bigger proteins but they're also being handled by much faster computers and certainly much faster up-/down-load speeds.

We need some new settings on both the client and the server, and unfortunately until those settings are adjusted, max-packet-size is effectively unsupported.

The reddit request makes sense to me, but it's up to the Pande Group to figure out the best way to respond, and discussing it here isn't an effective way to facilitate the best response from them.