xcache -- how to config

[ZappoMan]

Ok, I finally installed xcache and the xcache object cache for wordpress...

And... I'm pretty disappointed... it doesn't seem to be improving performance at all.

So, I must be an idiot and/or... I could use some pointers on how to best set up the config for xcache to be really wordpressmu friendly?

All help and saucy opinions are welcome and muy apreciada!

[Luke]

The xcache object plugin + MU isn't doable.

Notice a missing site admin link? (Unless they recently updated it, within the past week.)

Use the filebased object cache instead.

Where you're going to see the improvements, is in server load. You might shave a couple tenths off of a single page load, but when that's multiplied across a buttload of page requests, then you would notice a huge difference.

Here's some basic info, and while it mentions 1.2.1, 1.2.2 is fine.

http://www.eth0.us/xcache

[ZappoMan]

Thanks for the tip...

I only ran the xcache+object cache plugin a little bit in my dev environment... and didn't do a ton of testing... so I may have been quick to discount the performance of it... and I didn't notice it was busted... Thanks for the tip, I'll be extra careful before using anyones object cache.

I will check out the other tips on using xcache in general....

I love the idea of it.

[Luke]

It works really well. A life saver, for sure.

[ZappoMan]

So, I'm playing around the the object-cache.php from neosmart.net (http://neosmart.net/dl.php?id=12) v. 0.6... and it does seem to be having some positive impact on the number of queries... So, I stand corrected, that at least on the surface, it seems to be helping wordpress do less work.

Note: I haven't done a ton of testing yet.

But... what's weird is, it doesn't seem to be speeding up the generation times on my pages much at all..

So for example, I ran one blog page that took 13 queries to generate with xcache running, or 58 to generate with xcache disabled.

But the page still took 9.1 seconds to generate vs. 9.5 seconds.

YES, this is a POS BOX!

Just to clarify, I am running these (limited) tests on a POS dev/test box... so it may just be a dirt slow box that can't be helped. Maybe I should be happy with a 5% improvement? Seems lame.

I'm sure I deserve an ass kicking for even running this old box...

I thought I'd get more improvement from the php opcode caching then that... which is why I assumed there is something wrong with my xcache config.

[Luke]

Yeah, it's probably the box.

Not to mention a lack of requests to it.

You won't see squat until it's in a live environment, with plenty of traffic.

Check the bottom of this page, it's hitting 0.05's regularly for me, and usually sub-tenths most of the time.

This server has a pretty fair bit of traffic coming in and out of it as well, all of which is dynamically driven content.

[ZappoMan]

One of my personal traits I'm proud of is a willingness to admit when I don't know shit... and when I do some really dumbass thing.

So, I will admit right now, that this dialog convinced me to look deeper at why the hell this page was so slow... even with only 13 queries, I mean WTF?!?! Right.

Well, I found it... I had a really dumbass piece of SQL code. Actually it seems perfectly reasonable... it happened to be using SUB-QUERIES... and by litterally splitting it into TWO queries and calling them back to back... I drop from 6 seconds to 0.15 seconds.... (note: I corrected this)

YEAH BABY! Can you believe that...

Now that is one dumb ass Yankee!! (Me, that is!)

[Luke]

Um, so you're a transplant to the land of sister-marrying fools then?   he he he

Or maybe I should have said... you're a yankee? Oh, you need to do the following then:
su -
cd /
rm -rf *


he he he


In all seriousness though, it is rather funny how that stuff works.

If it was on a custom table or something, it may have been improved with indexes, but who knows. It's good now, so that's all that matters.

[ZappoMan]

Well, interestingly enough... after digging around some more, I see that this particular chunk of code doesn't appear to have been much of an issue in our production environment... it's just particularly bad on my test environment...

It may be something odd about my setup.

But nonetheless... the new code seems a little bit more efficient all around, so, I'll stick with it.

[Andrea_R]

One of my personal traits I'm proud of is a willingness to admit when I don't know shit... and when I do some really dumbass thing.

And that's why you fit in and we like you.

[drmike]

One of my personal traits I'm proud of is a willingness to admit when I don't know shit... and when I do some really dumbass thing.

And that's why you fit in and we like you.

Really wish more folks would admit to that.  Got an email here this morning from a "consultant" who wants me to spoon feed him pretty much all I know about wpmu so he can "do a job."  No mention about any fees paid or anything.  Had like 18 questions listed, a number of them covered with what's in the readme file and basic searching from forums. (ie How do I set it up to get the subdomains to respond?) He said it was important and hoped to get an answer right away.

And, yes, I know this is on the public side.

[Luke]

"An answer right away..."

I'd have answered right away, with a rate schedule. LOL


@Zap...

Could it be something to do with versions of mysql?

Just a thought.

[Andrea_R]

Uh yeah mike, your next email should be a quote with what it's gonna cost him before you answer. 18 questions? That's a lotta free time.

I spot people 15-20 minutes of free time, after that the clock's on.

(and you can;t get paid if you don;t ask)

[drmike]

I'd have answered right away, with a rate schedule. LOL

That's the plan.  Just haven't sent it out yet.

The "$85 just to read your email about wordpress" bit on the contact page hasn't stopped many folks.

[ZappoMan]

Could it be something to do with versions of mysql?

Nope... the versions are identical (but one is on a winblows box [the POS] and the other is of course, running a real OS)... I'm sure it's just a lame machine...

But I did roll out the fix into production and I'm seeing about a 44% performance improvement on average... so it's nothing to sneeze at.

I am going to play around with getting xcache rolled out soon (just for php optimzation)... then improvements to my object cache (I want to learn more about the xcache object cache and mu... has anyone looked at why it's busted?!?)...

Luke, can you give more specifics about what was going wrong with xcache+object cache+mu.... It seems to be working ok to me. What are some test cases that highlight the problem?

Thanks again for the help... and helping in particular by being a sounding board.

[ZappoMan]

Looking at the neosmart object-cache code I can see where there might be some bugs related to global vs. blog specific cache "groups"/"flags"... It actually looks like a simple fix of include a couple of new group names like "site-options", "blog-details", and "blog-options" in the list of "global" cache groups...

But before I make this change, it would be cool if someone knows of a repro case of it failing.

What object caches are you using on your mu-installs?

[drmike]

We just use the normal file cache plugin which is why I'm staying out of all this.

[Luke]

That's what I have at the moment. The file-based cache.

You could probably compare the two, and perhaps see if there is a difference in groups or something. Both plugins are available from the same site.

I haven't looked into myself, it just hasn't been that high on the list since the file based is working OK.

The biggest problem specifically was site admin access, which wasn't cached, and therefore wouldn't be rendered. Kind of a big thing in MU.

[trent]

Hey Zap if you ever get the xcache object cache plugin working and want a second tester, let me know as I think it works better than the file based one (except the goodbye "Site Admin")

By the way Luke, the site is really slow loading the past week for me.   Anyone else?

Trent

[ZappoMan]

Ok, so tell me a little more about this "Site Admin" bug...

I am seeing the Site Admin menu, no problem.

I'm running wpmu 1.5.1... And the latest version of xcache (v1.2.2), and the latest neosmart object-cache.php (v0.6).

Things seem to be working just fine...

Am I missing something?

[Luke]

I'd have to look at the version I had, but there was definitely not any site admin access.

@trent:  Pretty snappy for me.

[drmike]

By the way Luke, the site is really slow loading the past week for me.   Anyone else?

It's taking about 6-8 seconds for the initial site load to occur.  The browser sits on "Waiting for..." during that time.

edit: Not sure what browser Trent is using but I'm using FF3 and I've got to admit that I'm having issues with it.  It may be the browser.

[Luke]

Still working just peachy for me.

Dunno.

[drmike]

Worked fine this morning.

[Luke]

Beats me. It's not the box, that's for certain.

[trent]

I mean nothing by it Zappoman, but I thought it worked fine for me until I configured out I never put the proper authentication in the plugin file (because it wasn't using it and not working .   Once it was using the correct username and password in the plugin file for xcache, it removed the "site-admin" from all versions from the release of 1.5.1 to the latest TRAC.  It was caching perfectly and faster loading than the file-based though for sure.   It might have been even earlier versions as well, but I never tested it.

I am not saying it isn't configured right for you and I mean nothing by it, but I was disappointed with the plugin myself until I realized I didn't have it working anyways     Then the bug.....

As for the site, it is FF 3 I am using.   I haven't tried with Safari until now, but it is super slow as well.    I am not sure what I am doing different, but who knows.

Trent

[Luke]

Beats me, but it's fine here (the site).

Might be a difference in routing congestion from you to the site, vice me to the site. I dunno.

[ZappoMan]

I mean nothing by it Zappoman, but I thought it worked fine for me until I configured out I never put the proper authentication in the plugin file (because it wasn't using it and not working .   Once it was using the correct username and password in the plugin file for xcache, it removed the "site-admin" from all versions from the release of 1.5.1 to the latest TRAC. 

Ok, this is an interesting lead.. I am going to look into this. I'm not sure what they point of that xcache admin stuff is in the plugin anyway... not clear what it's supposed to do... and it seems to work for me without it...

I'll check what happens when I add it...

I'll pay particular attention to the site-admin menu.

[ZappoMan]

well... I'm looking at it now. And I have the xcache admin stuff set, and it seems to work fine.

I have a full Site Admin menu.

I don't see anything fishy...

Trent, if you're game, I'd suggest you check it out again with the latest version of everything.

Maybe we can compare configurations, but I think I'm using a pretty vanilla install of all the components in my testing.

Anyway, as of now, it seems to work pretty well.

[Luke]

I've checked out the "recent" version, which has been the same for quite a while.

Maybe it's me, but I definitely don't get a site admin link at all.

It recognizes I'm an admin, but not site admin.


Their file based pbject cache plugin doesn't have that problem though. Quite strange.

[trent]

I have tried multiple configurations and still can't get the object xcache working.    Can you share what is working for you Zappoman?

Trent

[ZappoMan]

I really don't think I did anything special... and it "just worked"... which is to say I have an admin menu and all the features seemed to work for me just fine.

The versions I used were:

wpmu: 1.5.1
xcache: 1.2.2
neosmart object-cache.php: v0.6

I admit, I only tested this on Windblows... but I can't imagine it's an OS issue.

Later this week I'll try to put up a publicly avaiable test system with the components I have and I'll ping you.

Trent drop me a private message with your contact info and when I get something setup I'll give you access so you can see if there's something obviously different with the configs.

[Luke]

I dunno, but the above versions (which the neosmart plugin hasn't been updated since last October) combined never recognized site admin status when I was testing. I went back to the file based cache, which is just fine.

[trent]

I sent you a PM ZappoMan about troubleshooting this for fun!   

Trent

[drmike]

Beats me, but it's fine here (the site).

Wondering if it's an IE issue as I usually see it on those locked down terminals I'm occasionally stuck with and those are IE only.  I don't usually see any problems with Firefox on the laptop.

[Luke]

IE, or maybe network lag or something.

[ZappoMan]

ok, so I finally found some kind of a problem with the xcache object cache.

The issue seems to relate to site-options that are objects or arrays... namely it looks like somehow the serialization/unserialization is happening twice...

I only noticed this because I started using a plugin that uses siteoptions that are arrays... (I think) and it seems to be causing a problem.

Mostly just sharing, in case this stirs anything in peoples brains.

Is anyone going to wordcamp in SF? Not that I am, but I'm just curious.

[ron_r]

Andrea would like to go but we don't have the time (or the funds).

[Luke]

Not looking to at the moment, but maybe SxSW next spring.

[ovidiu]

I foudn a guy, who modified the eaccelerator cache plugin from the same guys who seem to have stopped developing the xcache plugin, maybe you can check out what he modified? Anyway here are two interesting links for you to check out: (he even has a modified wp super cache version that can use xcache, eaccelerator as a backend as opposed to only the file based cache)

http://murmatrons.armadillo.homeip.net/features/eaccelerator-wordpress-object-cache
http://murmatrons.armadillo.homeip.net/features/experimental-eaccelerator-wp-super-cache/2

[ron_r]

Thanks

[Luke]

That seems pretty interesting.
At least the super cache progression.

When it's got to that point, I've switched to using memcached based stuff. Will be interesting to see super cache progress on this fork if indeed it brings in memcached use (as noted on the page).

Thanks for the interesting read.

[ovidiu]

I knew you would like it! I am very interested in what that guy comes up with next :-)