May 7
Some universities use powerful computers for a number of problems beyond the reasonable capacity of a normal computer. Is it more economical to use these computers than use a distributed computing system to make use of the spare capacity of their many, many desktop computers? --129.215.47.59 (talk) 12:26, 7 May 2014 (UTC)[reply]
- It depends on the problem - in particular how linked the various bits are. One can easily search for primes using distributed computing because the only thing each processor needs to know is which ones it should check. It would be very difficult to get any useful speed increase using distributed processing in weather forecasting because that depends on what happens all around the world. Plus of course you need a lot of people to be interested in the distributed computing problem. Dmcq (talk) 14:09, 7 May 2014 (UTC)[reply]
- Some supercomputers are essentially performing distributed computation. As Dmcq says, what is optimal depends on the nature of the problem, and how parallelizable it is. Automatic parallelization attempts to parallelize problems algorithmically, but traditionally it takes lots of human cleverness, unless the problem "naturally" consists of independent sub units, e.g. the prime hunt mentioned above. (As a small point of fact, global circulation models are usually run on large clusters, though computer scientists and physicists have spent huge amounts of effort to allow for it.)SemanticMantis (talk) 14:51, 7 May 2014 (UTC)[reply]
- I would say that most university supercomputers are massive clusters. They may have a high-speed interconnect, but it has gotten to the point where gigabit Ethernet is enough for most applications. A dedicated cluster is preferable to using the other on-campus computers as compute nodes because of the level of control and reliability that you get from dedicated systems. There's no risk of the PC's other users rebooting or shutting down computers, the OS and performance is consistent across nodes, and the system as a whole is simpler to reconfigure. Dedicated ethernet switches mean that you can expect consistent transfer rates between nodes. It also allows better heat management - a coworker and I tried installing some grid computing software in a university computer lab once, and the CPUs running at 100% 24/7 overwhelmed the building's HVAC system. Katie R (talk) 19:03, 8 May 2014 (UTC)[reply]
Over at WP:AN, there's a discussion about User:10.68.16.31, which has been making lots of bot-type edits; it's clearly User:ClueBot III editing while logged out. However, WHOIS and Geolocate both tell me that 10.68.16.31 is a private IP address. How can an address from a private network be assigned in a way that it's credited with edits? I've heard of people being able to spoof IP addresses, e.g. to avoid being blocked for stuff like this, so I can imagine someone spoofing a private address — but obviously that's not the case with one of the ClueBots. Nyttend (talk) 22:29, 7 May 2014 (UTC)[reply]
- Yes, it's running on a machine inside WMF's private network. "Private" means "not routable across the public Internet; in this case it's going from one WMF machine to another, inside that private network. The MediaWiki software just sees the address of the client (user agent) that's talking to it, and reports that. -- Finlay McWalterᚠTalk 22:40, 7 May 2014 (UTC)[reply]
- [edit conflict with 82.44.76.14] So the ClueBots are running on WMF computers; I didn't know that. I just figured they were run on the ClueNet computers or on Cobi's (or someone else's) private home computer. If someone started editing from WMF-run computers, would we likewise be shown a private-network IP address? Six months ago, when localhost was editing due to a glitch, someone suggested that people were editing from the server computer. Nyttend (talk) 23:01, 7 May 2014 (UTC)[reply]
Here's an interesting case 82.44.76.14 (talk) 22:57, 7 May 2014 (UTC)[reply]
- I think that those 10.* bots are running from toolserver. They shouldn't be allowed to edit while logged out, since it loses the info that the edits came from a bot. That's funny about 127.0.0.1. There is also one from 8.8.8.8 (the Google public DNS server). 70.36.142.114 (talk) 02:42, 8 May 2014 (UTC)[reply]
- According to https://wikitech.wikimedia.org/wiki/IP_addresses , 10.68.16.xx is assigned to machines in WMF's Ashburn, Virginia cluster ("equiad"). Cluebot migrated from WM-Deutschland's Toolserver to WM-Labs "Tool Labs" at equiad in March (ref). A list of the equiad machines is here, but I'm not aware of a detailed IP map, so I don't know which one(s) specifically host Tool Labs. -- Finlay McWalterᚠTalk 08:27, 8 May 2014 (UTC)[reply]