Hey rocky, watch me pull time out of my hat...

What rolls down T3s, boosts connect fees, and makes your throughput drag? Makes it tough to hack, won't get off your back, it's lag lag lag!

Okay so we've all noticed some lag, fine, whatever. On a seperate note some of us have noticed that some people, those who should know better (me too, me too!), occasionally reply to nodes without okaying them. I know one reason why I use the consider/edit nodelet less than I might (including for okay), is that it is a seperate form and requires an additional submission. So what if it were (optionally?) inlined in the page body as a part of the normal form? This way one could okay and vote or okay and reply at the same time.

Another thing to consider is cleaning up page markup a bit. I was able to eliminate 1.6k of 13.2k by removing unnecessary whitespace. I'm sure there are other things that could be tidied up as well, though probably for more modest gains. And some of it could be done automagically e.g. Apache::Clean. There is also the possibility of sending content encoded e.g. Apache::Compress, to clients that can receive it.

Yes, from our end we cannot tell if the lag is due to bandwidth or other forms of throughput issues (disk/CPU), so much of this is a shot in the dark. I do realize that all but the first proposal costs CPU to gain bandwidth (and the associated, probably lesser, CPU gain). Unless the server is already pegged it's probaby woth some experimentation though.

-- perl -pe "s/\b;([st])/'\1/mg"

2002-02-18 Corion : chop; chop;

Comment on Hey rocky, watch me pull time out of my hat...

Replies are listed 'Best First'.
Re (tilly) 1: Hey rocky, watch me pull time out of my hat... by tilly (Archbishop) on Feb 18, 2002 at 18:56 UTC
Without access to logs and debugging you really can't tell anything about where the problem is likely to be. For instance one that I was part of trying to track down was finally tracked down last week to Linux machines being set to try to communicate at 10 or 100 on an ethernet, while the Cisco router was set to 100. Most of the time it worked, but occasionally it messed up, causing a dropped connection and a socket hanging for 5 minutes (which is what Apache's timeout was set to). Configure the Linux machines to only try to talk at speed 100 and the network problem went away. Would you have ever thought of that? I wouldn't have. The person who figured it out started in the system logs and noticed a lot of TCP/IP errors, then went from there...	[reply]
Re: Re (tilly) 1: Hey rocky, watch me pull time out of my hat... by joealba (Hermit) on Feb 19, 2002 at 05:25 UTC
Yeah, I would have thought of that, tilly.. :) But, only because something like that just happened to me too! Cheap 10/100 autosensing hub + autonegotiating 10/100 switch on a 10 MB/s network = Collision Fest 2002. Regarding the lag -- I am certain that page size is NOT the issue. It's the generation of the page that's the issue. The site is completely dynamic. The site works great for me at night (EST), but chunks at other times -- generally when there are lots of monks active. So, there's a bound somewhere -- most directly caused either by hardware or code. The admins & coding monkeys know that there's a scalability problem. They have the logs, the access, and the knowledge -- They'll fix it. I have faith in that. (joealba trails off into a perlmonks "I have a dream..." speech preaching about a faster site) I'm sure they know that any one of the monks would be glad to donate their time to review code for bottlenecks, myself included.	[reply]
Re: Re (tilly) 1: Hey rocky, watch me pull time out of my hat... by belg4mit (Prior) on Feb 18, 2002 at 19:08 UTC
Exactly, that's along the lines of what I was alluding too (even if I wasn't explicit) in the last paragraph. And for the record, no that wouldn't have occured to me while debugging. Although when setting it up I probably would have explicitly set it to 100 to match :-D. And the other thing I was trying to allude to was there are some things that are just generally helpful in the long run. `-- perl -pe "s/\b;([st])/'\1/mg"`	[reply]
Re: Hey rocky, watch me pull time out of my hat... by belg4mit (Prior) on Feb 19, 2002 at 05:55 UTC
I might try to submit a patch for this once I get setup with e2. But the idea being that the same functionality can be acheived without moving the nodelet. Since as far as I can tell all forms (on note, perlquestion, etc. types) have their action set to the workhorse index.pl and method=POST we can make the whole page live in one form. It seems index.pl is already setup to act based upon "op", and instead of it being a hidden field the submit buttons could simply posess this name. There might be some other issues with other hidden fields e.g. different nodelets having fields w/ the same name but different values but I've seen nothing yet. Added bonus for simplifying the HTML. Just think, chat, vote, and approve in one fell swoop :-D. Actually though one must explicitly look for combined actions, so potentially confusing combinations need not be supported. In fact probably only approval needs to be processed when vote or submit (from a reply) is hit. Of course this would mean sending some unused data if you don't do everything everytime, very little though. Oh know it all sounds rather complicated but there really isn't much of anything to it. This is just one way of implementing it (instead of the much easier wholesale movement of the approval nodelet) since there are some who would not like to see special cases for display brought into the body nodes themselves. Which is more favorable? `-- perl -pe "s/\b;([st])/'\1/mg"`	[reply]
Re: Hey rocky, watch me pull time out of my hat... by ignatz (Vicar) on Feb 19, 2002 at 15:12 UTC
This may seem blasphemous, but before trying to fix the problem of generating the page with all the its current functionality, wouldn't it be better to eliminate functionality that isn't core and drags down the generation of the page? ()-() \"/ `	[reply]
Re: Hey rocky, watch me pull time out of my hat... by belg4mit (Prior) on Feb 19, 2002 at 07:44 UTC
And finally (honest!) probably the easiest to implement, a "submit & comment on" button added to the approval nodelet. Which upon submission displays a comment submission form. Helas I'd probably go for this one myself (as fun as the others might be ;-). UPDATE: Currently to do what is proper one must; click okay, hit submit, wait, click comment, enter comment. This would change the procedure to; click okay, hit "submit & comment on", enter comment. It amounts to a redirect because a different button was used. Social engineering + convenience. `-- perl -pe "s/\b;([st])/'\1/mg"`	[reply]