To answer the OP Question.
As static files, directly via disk thru web server.
at only 100px they are ideally suited to in memory caching by the Web server.
There is a plethora of info ,caching strategies, configs, how-to's for just about every web server out there.
Infact - The best option in terms of user experience (the image speed you refer to) is to use a CDN capable object store. period.
The "DB" as static storage choice is simply expensive - in terms of all the overhead processing, the burden on the DB, as well as financially, and in terms of technical debt.
A few things, from several answers
Google images and Yahoo images are using base64 and serving images
inline. Check source code and you'll see it.
No. They absolutely do NOT. Images are mostly served from a static file "web server" Specfically gstatic.com:
e.g. https://ssl.gstatic.com/gb/images/p1_2446527d.png
compact - you have single file. also if file is compressed, base64
image is compressed almost to the size of normal image.
So actually, No advantage at all, plus the processing needed to compress?
page is retrieved in single request.
Again, multiple parallel requests as opposed to a single larger load.
What happens when 100000 users request that 200 images on a very
loaded site. The disks of the server, returning the images should
collapse.
You will still be sending The same amount of data, but having a Longer connection time, as well as stressing your database. Secondly the odds of a run of the mill site having 100000 concurrent connections... and even if so, if you are running this all of a single server you are a foolish admin.
By storing the images - binary blobs or base64 in the DB, all you are doing it adding huge overhead to the DB. Either, you have masses and masses of RAM, or your query via the DB will come off the disk anyway.
And, if you DID have such unlimited RAM, then serving the bin images off a Ramdisk - ideally via an alternative dedicated, lightweight webserver static file & caching optimised, configured on a subdomain, would be the fastest, lightest load possible!
Forward planning? You can only scale up so far, and scaling a DB is expensive (relatively speaking). Again the disks you say will "sp
In such a case, where you are serving 100's of images to 100000 concurrent users, the serving of you images should be the domain of CDN Object store.