[Seaside] Seaside for large, high-traffic sites.

[UZONYI Levente]

Hi!

On Fri, 11 Sep 2009, Joseph Harrison wrote:

> I have some questions about Seaside scalability.
>
> In 2008 I wrote a small webapp in Seaside and overall found the experience very
> enjoyable; so much so that I would prefer to do my next project in it.
> Unfortunately, due to its nature, this project will receive considerably more
> traffic than the last, and I am somewhat skeptical of Seaside's (and Squeak's)
> ability to scale.
>
> I understand Smalltalks (especially Squeak) use green threads for concurrency,
> so obviously a single VM process handling all of those HTTP connections would
> not work. Rather, the solution seems to be some sort of parallel setup with many
> VM processes running simultaneously, each using green threads as needed, with
> some load balancer in front of them. Do they share the same image (in which
> case, I am guessing they can't save the image), or must each have its own? Next,
> persistence: I need a robust alternative. It need not be a pure object database,
> just something with decent performance that can scale; a mature Squeak interface
> to an RDBMS like PostgreSQL would suffice. Additionally, I am curious about
> 64-bit Squeak and if it can work with Seaside.
>

We are using Squeak with Seaside (2.8) and Postgresql as backend for a 
site. We got an 8 core (1.86GHz Xeon) server running Apache, (at most) 8 
images and Postgres. The images can't save since they are using the same 
image and changes file (but we don't want to save at all). A single Squeak 
image with Seaside can handle about 75-100 'hello world' requests/second. 
With this setup we measured 160-200 requests/second for an average page 
(with data fetched from database).

We are using a modified version of the PostgresV2 package which gives us 
extra speed and features (we also have a partial implementation of the v3 
protocol which we are about to test in another project).

We found that 64-bit vms are slower than 32-bit versions, so we are using 
32-bit vms with ia32-libs.

Failover isn't that good with this setup since if an image crashes, all 
users of that image lose their session data (sticky sessions). Storing 
state in a datastore could help, but we never had an image crash.

> Am I setting myself up for a lot of pain, or is it really practical to use these
> tools with large websites? If Squeak + Seaside isn't enough, how hard is it to
> drop down into C for extra performance or to call out to foreign code?

FFI is pretty easy, writing plugins are a bit harder, but i wouldn't 
bother with this. Scaling Seaside horizontally is much easier.

Cheers,
Levente

>
> Thank you in advance.
>
> _______________________________________________
> seaside mailing list
> seaside at lists.squeakfoundation.org
> http://lists.squeakfoundation.org/cgi-bin/mailman/listinfo/seaside
>