My forum has gone blank and I only see a white index page

[Arantor]

Except you're talking about FAT entries. Modern servers don't generally have the classical FAT of old. NTFS and ext2/3/4 certainly don't quite work that way.

Also, you're assuming the only kind of caching is file based. It's just the one most users can actually use, and better to hit a ready made chunk of data that can be frequently polled and stored in that page file of yours rather than making considerably more I/O to the database to get the same data. I guarantee you the sum total processing of all types to get from the cache is less than getting the same data without a cache.

This of course ignores the caching done by your CPU, your operating system, your environment and your database server.

IMO you are giving people bad advice, and I'm so glad you don't handle the infrastructure I work with professionally where I scale to tens of thousands of hits per minute, which is only possible by way of caching or VASTLY more hardware than my clients can afford.

[GigaWatt]

Except you're talking about FAT entries. Modern servers don't generally have the classical FAT of old. NTFS and ext2/3/4 certainly don't quite work that way.

Nevertheless, they still have to make an entry where that file actually physically resides. With large files, that's usually not a problem. Divide the same file in a million different chunks, see what happens.

Also, you're assuming the only kind of caching is file based. It's just the one most users can actually use, and better to hit a ready made chunk of data that can be frequently polled and stored in that page file of yours rather than making considerably more I/O to the database to get the same data. I guarantee you the sum total processing of all types to get from the cache is less than getting the same data without a cache.

If the cache file is large enough, yes, it will benefit performance. As I said previously, I approve of caching only in page files. If the same principle can be applied without having millions of small files, I'm would give it a try and it will probably be beneficial.

IMO you are giving people bad advice, and I'm so glad you don't handle the infrastructure I work with professionally where I scale to tens of thousands of hits per minute, which is only possible by way of caching or VASTLY more hardware than my clients can afford.

I would suggest to buy better hardware... but that's just me. If they can't afford it, it's their problem.

[Arantor]

So you're genuinely in favour of doing the same workload (of CPU, RAM, network AND I/O) over and over and over and over and over, rather than doing it once and reusing it (with a MUCH lower usage of all of the above) because you have an ideological problem with one specific implementation of the concept. Good for you.

[GigaWatt]

Yes, that's correct, that's what CPU power, RAM usage and bandwidth usage is for - to be used, even if that means doing things over and over... which would also be the case when powering up our PCs or laptops... they power on and off over and over, doing the same things for years, but I presume you wouldn't call that "doing things over and over".

And it's not an ideological problem if that problem increases my workload. Small files tend to get corrupt, which means I have more work on my hands.

[Aleksi "Lex" Kilpinen]

I'm gonna have to respectfully disagree - on a larger forum like mine ( around 3M posts ) the caching can do a huge difference during peak hours.
I have also never seen an issue with cache files getting corrupt with 2.*, apart from times when I am messing about with the code myself.

[Arantor]

You're still missing my point.

A hundred people visit the board index at once. 90 of them are guests, for those 90 people you could run the multiple queries required to build the board index each time, or you could run the set of queries once and reuse the result. This is every case more efficient meaning that you can use the resources you have to do more for the same price. You seem to have issues with the notion of using resources efficiently.

That said, you seem to be opposed to the very idea rather than the one specific implementation that's an issue, because memcache doesn't persist this to disk, neither does Redis, nor APCu, but you're against the entire concept of caching rather than the one type you disagree with, because you had issues with it in the past.

Moreover, you assert that it creates more work for you, which is also odd. Small files do not inherently get corrupted any more than big ones. Highly contested files of any size do, on the other hand, get corrupted more frequently than non highly contested files.

As evidence of that, have a look at all the poor souls whose database has crashed, these typically happen to larger files and all because of high contest over the files, not because small files are somehow more at risk than large ones.

[GigaWatt]

Why don't we just agree that we disagree . Is that OK ?

[Arantor]

No, it's not OK, actually. It'd be like saying 'why don't we agree to disagree that the earth isn't really round' when all available evidence shows quite clearly that it is. You're entitled to your opinion, sure, but when you're going to use that opinion to spread misinformation that people will then trust because you have a badge, I have a problem with that.

[GigaWatt]

It's only misinformation if you can prove it. You can't prove that disks don't have problems with small or large files (in theory, the shouldn't), and I can't prove that they do. I can only say this from personal experience and info from various sites I've gathered over the years.

So, to clear things up.

Everything I wrote previously about caching is my own personal opinion based on my personal experience. There is nothing wrong with having cache enabled. It's my personal view that it should be disabled, but that doesn't mean that mine is the only opinion that counts.

Oh, and another thing.

My advice would be to disable it.

It's advice. The OP can either take it or not. It's an option, it could either stay enabled or disabled. There is nothing misinforming about it.

[Arantor]

The only evidence I can find for performance degradation with many small files is when you start getting into tens of thousands of files. Some file systems (NTFS for example) will actually even inline the file into the file table directly if it's small enough.

But you're not interested in being open minded to things outside your knowledge base, focusing on the one part of it you think you know and getting defensive about it, so we're done here.

[Kindred]

My advice would be to disable it.

It's advice. The OP can either take it or not. It's an option, it could either stay enabled or disabled. There is nothing misinforming about it.

However, in this case, it is advice based on an opinion based on incorrect or incomplete data.

While everyone is entitled to an opinion, facts trump opinion when it comes to information... and while you may choose to maintain an opinion based on incorrect/untrue facts, that does need to be highlighted when you share that opinion.