[Koha] How to measure and improve Koha performance for our instance?

[Mike Lake]

Hi all

On 2022-01-17 5:16 pm, dcook at prosentient.com.au wrote:
> Great to see another Australian library using Koha!

Yes. We are pleased to be able to use such nice NZ Open Source software.

> It looks like you're running Debian Buster (based on your Apache
> response), so I'm guessing you used the Debian packages to install
> Koha? I thought that new instances created this way used Plack out of
> the box, but I think I might be mistaken...

Initial install was Debian packages onto Stretch in 2019. Then an 
upgrade from Stretch to Buster in September 2021 so thats prob why Plack 
is not installed.

> If you enable Plack/Starman, page loads and search times will be
> faster, because the application code is run in persistent processes
> that use pre-loading and caching of code. However, that comes at a
> cost. In practice, I find each Plack/Starman process reserves about
> 200-500MB RAM over its lifetime (default 50 requests before it's
> recycled if I recall correctly), and the default Koha configuration
> uses 2 Plack/Starman processes. Depending on expected usage, you may
> want more/fewer Plack/Starman processes. (Note that those processes
> are shared across the Staff Interface and the OPAC.) Overall, enabling
> Plack is the best way to improve Koha performance.

> (Another thing to note is
> startup for Plack/Starman processes are CPU resource heavy as a lot of
> work goes into setting up and verifying the REST API.)

> Overall, I'd say the more CPUs the better, especially if you're
> running your database on the same server.

Yes running on just the one server.

Excellent info above, so useful to know this.
 From that perhaps its better to double what I have now to 2 CPUs and 4 
GB RAM and then look at adding in plack for better performance again 
rather than adding in plack now and later doubling CPU and RAM.

> I think Memcached might technically be optional, but you'll want to
> keep it for performance. Without it, performance would certainly
> degrade.

OK.

> In terms of metrics, you can use third-party log analyzers (for the
> Apache logs) and web analytics (eg Google Analytics) to get an
> overview of how the application is performing overall. For more
> targeted analysis, I tend to just use the Developer Tools in the
> browser. For instance, I see your OPAC is taking about 4 seconds to
> serve the homepage. I tried one of the instances I manage (on a high
> specced server using Plack) and it took 333ms. There are other tools
> you can use for profiling the application, but that's a something you
> might want to explore with the "koha-devel" list instead.

Ah! I had forgotten about using developer tools. I have those in my 
browser and yes I see there is a performance tab which shows lots of 
load times for components.

> More reading:
> https://lists.katipo.co.nz/public/koha/2020-December/055598.html
This said: "I'd say minimum 4 CPU between all the various different Koha 
subcomponents for any serious usage"

For us I can go to 2 CPUs for now but 4 is a bit too pricy. We can prob 
go to 4 when we can show how much the library is being used later on :-)

> I haven't used Linode so I can't really comment on its features, but
> other platforms like AWS make it very easy to switch between different
> CPU and RAM sizes.

Pretty easy. I use AWS also. Going up in a plan is easy as disk is 
expanded but going down can be a problem if your disk usage no longer 
fits in the smaller plan.

> I'd say your first step is to enable Plack (on your test instance),
> and just manually compare similar web requests (like loading the home
> page, performing the same catalogue search). You should already have
> the RAM to handle it. I could see you getting into strife only having
> 1 CPU, as that 1 CPU would have a lot of work to do, but that's where
> the log analysis / web analytics can help you see real life usage.

First step tonight will be to read up on configuring and turning on 
plack for the test instance. Then I'll toss up doubling cpu/ram before 
further tests.

> I hope that helps. Feel free to send more questions.

Thanks. Much appreciated.


> -----Original Message-----
> Date: Fri, 14 Jan 2022 16:04:11 +1100
> From: Mike Lake <mikel at speleonics.com.au>
> To: Koha <koha at lists.katipo.co.nz>
> Subject: [Koha] How to measure and improve Koha performance for our
> 	instance?
> Message-ID: <daad1c0c667cacb91cb604a2e7797a9c at speleonics.com.au>
> Content-Type: text/plain; charset=US-ASCII; format=flowed
> 
> Hi all
> 
> We have a production library setup at https://opac.caves.org.au and
> its now ready for users to use. We also have a test instance that can
> be used for testing. I'm wanting to know the options for measuring and
> improving the performance - just basic load and search time that users
> perceive.
> 
> Currently we are running on a Linode instance with 1 CPU and 2 GB RAM.
> Yes it's a small library :-) Looking at Linode Longview it shows we
> have not hit more than 7% CPU usage and it usually sits at 1% CPU and
> 670 MB mem when idle. During a query cpu goes to 100% for a short
> period and mem to ~ 1 GB.
> 
> I see in htop that several memcache threads are running. That's now
> installed automatically with Koha? The wiki seems to suggest that its
> optional. And there is plack mentioned in the wiki. Should we be
> running plack?
> 
> I can double our Linode instance to 2 CPUs and 4 GB RAM but that's
> possibly an irreversible change (and double the $) so wish to leave
> that until I know it would make a significant difference. And how
> could I quantify that performance increase?
> 
> Thanks
> 
> --
> Mike

-- 
Mike