High volume sites

[georgew]

Hey there!

Has anyone tried to split the database and the php services across a cluster of machines?

I was told it was not possible, but I don't know why. Do we not properly maintain state with multiple web servers? Isn't that why you use a database instead of something faster?

I don't need it yet, but I'm expecting a couple million a day...
I need to know why this won't break.

Most acceses are reads, so I should even be able to mirror the database using non-realtime methods, and diret all users with writes to a master server. This will get around the speed penalty normally associated with database mirrors... So even if the software and the data MUST reside
on one machine I'm going to try and build somethig large enough to handle the load.

What are the busiest sites running geoclassifieds? If I still owned Auction.com I'd be demanding the same for geoauctions...


Thanks!
George

[michaelcastello]

georgew, you seem to have a lot of experience here. This is my first post so I may sound a bit novice, which I am, but I have many prime city domain names that I want to use this software on. I hope I don't need to get redundant with the software and licenses. So let's say that I wanted to put classifieds on Dallas.com, Atlanta.com and Boston.com and wanted to use this software on a site/server called Onion.com Classifieds. Could I have local visitors at Atlanta.com hit the classifieds link that brings them to the Onion.com Classifies / Atlanta page with local Atlanta classifieds results? Now let's do the same thing for 20 other cities. Can the software achieve this from one server/license/software? I hope this makes sense. Thank you for your time. Michael

[geomodules]

Hey there!

Has anyone tried to split the database and the php services across a cluster of machines?

I was told it was not possible, but I don't know why. Do we not properly maintain state with multiple web servers? Isn't that why you use a database instead of something faster?

I don't need it yet, but I'm expecting a couple million a day...
I need to know why this won't break.

Most acceses are reads, so I should even be able to mirror the database using non-realtime methods, and diret all users with writes to a master server. This will get around the speed penalty normally associated with database mirrors... So even if the software and the data MUST reside
on one machine I'm going to try and build somethig large enough to handle the load.

What are the busiest sites running geoclassifieds? If I still owned Auction.com I'd be demanding the same for geoauctions...


Thanks!
George

Its not designed for this..... might be possible but it would take and aweful lot of work to do.

[geomodules]

georgew, you seem to have a lot of experience here. This is my first post so I may sound a bit novice, which I am, but I have many prime city domain names that I want to use this software on. I hope I don't need to get redundant with the software and licenses. So let's say that I wanted to put classifieds on Dallas.com, Atlanta.com and Boston.com and wanted to use this software on a site/server called Onion.com Classifieds. Could I have local visitors at Atlanta.com hit the classifieds link that brings them to the Onion.com Classifies / Atlanta page with local Atlanta classifieds results? Now let's do the same thing for 20 other cities. Can the software achieve this from one server/license/software? I hope this makes sense. Thank you for your time. Michael

answered in other forum

[danzo]

I would thought to separate the database and the php script, set the database server to an internal ip address instead of localhost in the script config, and open the relevant ports.
If you use more advanced database like Oracle, I think you can replicate it. The script use adodb drivers though.

Hey there!

Has anyone tried to split the database and the php services across a cluster of machines?

I was told it was not possible, but I don't know why. Do we not properly maintain state with multiple web servers? Isn't that why you use a database instead of something faster?

I don't need it yet, but I'm expecting a couple million a day...
I need to know why this won't break.

Most acceses are reads, so I should even be able to mirror the database using non-realtime methods, and diret all users with writes to a master server. This will get around the speed penalty normally associated with database mirrors... So even if the software and the data MUST reside
on one machine I'm going to try and build somethig large enough to handle the load.

What are the busiest sites running geoclassifieds? If I still owned Auction.com I'd be demanding the same for geoauctions...


Thanks!
George

[dmc]

You could set up a round-robin IP redirect so each request get thrown to a different webserver. This would retain the connection for the duration on the front end. The Database could be manage by a clustered service (we do a lot of work with clustered Oracle (RAC) but it should be similar with mySQL.

HTH

[georgew]

georgew, you seem to have a lot of experience here. This is my first post so I may sound a bit novice, which I am, but I have many prime city domain names that I want to use this software on. I hope I don't need to get redundant with the software and licenses. So let's say that I wanted to put classifieds on Dallas.com, Atlanta.com and Boston.com and wanted to use this software on a site/server called Onion.com Classifieds. Could I have local visitors at Atlanta.com hit the classifieds link that brings them to the Onion.com Classifies / Atlanta page with local Atlanta classifieds results? Now let's do the same thing for 20 other cities. Can the software achieve this from one server/license/software? I hope this makes sense. Thank you for your time. Michael

Your answer makes perfect sense, it was the first question I asked the programmers before buying the product.

Are you with Boulevards?

I think the answer to your question is no, at least that's what the programmers told me. It might have changed since then.

I plan to use the zip-code-search to do this myself.

But I tell you, to compete with Craigs List and Ebay, we all need to band together somehow. Critical mass is an elusive thing... I'm working on a plan...

[georgew]

I would thought to separate the database and the php script, set the database server to an internal ip address instead of localhost in the script config, and open the relevant ports.
If you use more advanced database like Oracle, I think you can replicate it. The script use adodb drivers though.

Actually, I'm pretty sure that what you are saying will work... but that till leaves my question. Seperating the database from the web server will help. But not all web apps use the database for everything. That is the question...

If Geoclassifieds uses local variables for any state-keeping, then you cannot have more than one web server.

If the database is used for all variables, then you can have 200 web servers feeding off the same database.

When I asked the programmers this, they told me it was not possible to run the database on a second machine. Since I now how databases work, I know that this is not true.... But I also know that programmers don't always know everything about unix systems, especially when they program on windows machines.

As for database replication, that can be done with mysql, if you are running the right version.

But I suspect that Equineads is the one with the answer.... He writes modules for the system, so he may know it's limitations well...

But since databases can be the thing that slows everything down, it wouldn't suprise me if the programmers did not use it for everything needed to run multiple web servers... But that might be ok, a modern web server can serve a lot of web pages.

[dmc]

Hold a key in the URL grabbed from the session and put into a DB row. This links back to the DB which retains the state information. I use this a lot for many banking sites I write in the UK. No load on app servers RAM with heafty sessions and stateful (distributable) data.

BTW I was taught that eBay holds the key information described above in the HTTP header rather than the URL.