Anyone having performance issues with 5.6.3.1?

Permalink
I'm hitting some strange performance issues with 5.6.3.1, in particular the first load of a site if it hasn't been hit for a while.

When the load delay appears it's almost like it needs to load concrete5 into memory and that takes a while - once it's loaded up one page, other pages load as expected and without delay. Then perhaps half an hour later it will do it again. The initial load can be up to 20 before first byte.

I'm experiencing this with a bunch of new sites on a server I already have a handful of 5.6.2.1 sites on (that I obviously haven't updated yet). These sites are very responsive (2-3 seconds load max) and DON'T have this initial load issue. My caching settings are set the same across the site and I'm really not doing anything different between them. Even new sites with only 5 or 6 pages have this initial pause before loading.

I can even try loading up a 5.6.3.1 site, wait five seconds, open another tab and load up one of the 2.1 sites with tons of pages before the first even starts displaying.

I've tried using things like using memcache, apc, an even PHP5.5's opcache, but these only seem to adjust the normal load times, not fix this initial 'boot up' time I'm experiencing. It's not a matter of caching css/js/images either, as I'm taking about a delay of the initial page, not to complete rending a page.

The only temporary work around I've found so far is to set up a cron job to hit the site every half hour, to effectively keep it 'alive'.

So has anything significant happened to concrete5's caching lately that might be the cause of this? Anyone else experiencing this?
I was under the impression the changes made from 2.1 to 3.1 were really only bug fixes and small enhancements.

I even investigated if it's the automated jobs firing off incorrectly, but didn't find anything in that regard.

I'm really concerned about this.

mesuva
 
exchangecore replied on at Permalink Reply
exchangecore
Who are you hosting with? I haven't had any issues with any of my c5 sites, they are all consistently sending back the first page response in less than half a second. Have you tried comparing the same site on the same account with 2 different versions to ensure that it is truly the version different that's affecting it and not some other factor such as plugins or site content?
mesuva replied on at Permalink Reply
mesuva
I'm hosting with Digital Pacific, who I've always been very happy with.

It's a bit of a tricky one to diagnose as I don't have multiple versions of sites sitting ready to test at different versions. I did consider add-ons, etc, but one of the sites has no add-ons and is VERY simple in construction. I've also looked at autonavs in Stacks, but I've got plenty of sites with that setup that have no issues (even with big nav structures).

However I may have jumped the gun a little in thinking it's version specific, as I've just found a 5.6.2.1 site that is showing the same symptoms.. so it's perhaps a server issue, but I'm still confused why older sites are still performing well.. it could be they are simply receiving more traffic so they stay alive, whereas the newer sites aren't public yet as such.

As a side note though, I am finding that turning on the opcache and APCu for PHP5.5 is a speed improver.

Cheers
-Ryan
mhawke replied on at Permalink Reply
mhawke
On the slow site(s), check to see what the settings are for the AutoNav. Sometimes it's set to 'all', 'all' but the css is only set up to display 1 or two levels. The delay can be caused because the AutoNav has to load and inspect all the pages in your site even though you only render a few links.

Personally, I have not found any differences in the 'time to first byte' between 5.6.2 and 5.6.3 when waking up 'dormant' sites.
mesuva replied on at Permalink Reply
mesuva
I did think of this, but I'm definitely not fetching all pages and hiding them with css. On one site it's only 7 pages in total anyway.

I've tried both an autonav in the a Global area, as well as hardcoding it into the template, I think the hardcoding is slightly faster.

I guess I really just wanted to confirm it wasn't some bug with the latest version - I appreciate the feedback here.