Server Upgrade Advice Needed

[Dabbler]

OK. Everything's finally spelled out. (Really)

This is posted at the suggestion of Zachery,
based on my original post looking for a new host service:
http://www.vbulletin.com/forum/showthread.php?t=131859

================================================== ===
Here's where I am:

1 - dedicated server

2 - server specs (temp platform, after first server crashed)
Dual processor P4(?) 2.4 GHz (host insists they're not Xeons)
1 Gig RAM
63 GB RAID 5

OS: Windows 2000
IIS: 6.0
PHP: php 4.3.10 (just upgraded)
MySQL: v4.0.20a-nt
Server API: isapi (was cgi/fcgi)

3 - no innodb type databases

4 - mysql compiled/installed: See Post 3 below

5 - top stats
Threads: 45K
Posts: 515K
Members: 19K
Searches: 610k

Top Poster: 5,200 posts
Max member visits/day: 1,500
Max new threads/day: 350
Max posts/day: 2,800

Data Usage: 719.02 MB
Index Usage: 273.16 MB
Attachment Usage: 0.99 MB
Avatar Usage 0.00 MB
Profile Picture Usage 1.16 MB

MySQL Packet Size: 1.00 MB
PHP Max Post Size: 8M
PHP Memory Limit: None

6 - mysql configuration variables: See Post 15 below

7 - mysql extended-status output: See Post 5 below

8 - other scripts use mySQL: no

9 - concurrent users:
max: 200 (5 minute window)
We have a reported peak of 338 users. I wasn't online at the time,
but I think it was probably just a search engine spider.

I'd like to plan for 300 max users.

typical user count: 120

10- phpinfo.php:
http://206.173.31.52/phpinfo.php

11- Apache info: n/a

12- vB ver: 3.0.7

13- apache log files: n/a
================================================== ===

Right now forum growth appears to be hamstrung by our platform.
We are strictly a discussion forum. Minimal file transfers. Minimal graphic
uploads. (Using default my.ini; no mmcache; cgi/fcgi.)

With 50 people online (5 min win), I'm hitting 100% utilization
on the dual cpu's frequently.
Page delivery starts to slow.

With 75 people online (5 min win), cpu utilization is practically
pinned at 100%. (see Task Manager window in post 2 below).
Page delivery is noticeably slow. Sometimes 20+ seconds per page.
Features start to break.

We need to move off this temp server ASAP.

What should I be looking for?
Dual Zeon?
More RAM?
2 server setup?

Although the current OS is Win2000, we were on a Win2003 platform.
I believe I'd like to go to a linux solution (in order to get spiderable
archives).

Help and advice respectfully requested.

[Dabbler]

The attached task manager window shows memory and cpu utiliztion at 71 users (5 min win).

[Dabbler]

how was mysql compiled/installed?

Answer :
Before Installing Mysql, Install PHP first.
Then install mysql 4.0.20-nt (same version as we had with the original Forums Server).
The file is a complete installer that will configure itself.
According to the MySQL manual it's better to use binary distribution (With
Setup) than source distribrution. See details below

Installing MySQL on Windows

The installation process for MySQL on Windows has the following steps:

1. Obtain and install the distribution.
2. Set up an option file if necessary.
3. Select the server you want to use.
4. Start the server.
5. Assign passwords to the initial MySQL accounts.

MySQL for Windows is available in two distribution formats:

* The binary distribution contains a setup program that installs
everything you need so that you can start the server immediately.
* The source distribution contains all the code and support files for
building the executables using the VC++ 6.0 compiler.

Generally speaking, we used the binary distribution. It's simpler, and you need no additional tools to get MySQL up and running.

[Dabbler]

mysql configuration variables

See Post 15 below

[Dabbler]

your mysql extended-status output

Answer:
OUTPUT
Start
+--------------------------+------------+
| Variable_name | Value |
+--------------------------+------------+
| Aborted_clients | 135 |
| Aborted_connects | 61 |
| Bytes_received | 2477247875 |
| Bytes_sent | 1876060108 |
| Com_admin_commands | 0 |
| Com_alter_table | 0 |
| Com_analyze | 0 |
| Com_backup_table | 0 |
| Com_begin | 0 |
| Com_change_db | 544062 |
| Com_change_master | 0 |
| Com_check | 168 |
| Com_commit | 0 |
| Com_create_db | 0 |
| Com_create_function | 0 |
| Com_create_index | 0 |
| Com_create_table | 0 |
| Com_delete | 33722 |
| Com_delete_multi | 0 |
| Com_drop_db | 0 |
| Com_drop_function | 0 |
| Com_drop_index | 0 |
| Com_drop_table | 0 |
| Com_flush | 0 |
| Com_grant | 7 |
| Com_ha_close | 0 |
| Com_ha_open | 0 |
| Com_ha_read | 0 |
| Com_insert | 286606 |
| Com_insert_select | 3323 |
| Com_kill | 0 |
| Com_load | 0 |
| Com_load_master_data | 0 |
| Com_load_master_table | 0 |
| Com_lock_tables | 0 |
| Com_optimize | 167 |
| Com_purge | 0 |
| Com_rename_table | 0 |
| Com_repair | 2 |
| Com_replace | 30017 |
| Com_replace_select | 0 |
| Com_reset | 0 |
| Com_restore_table | 0 |
| Com_revoke | 0 |
| Com_rollback | 0 |
| Com_savepoint | 0 |
| Com_select | 5799085 |
| Com_set_option | 3 |
| Com_show_binlog_events | 0 |
| Com_show_binlogs | 8 |
| Com_show_create | 0 |
| Com_show_databases | 38 |
| Com_show_fields | 0 |
| Com_show_grants | 1 |
| Com_show_keys | 0 |
| Com_show_logs | 0 |
| Com_show_master_status | 0 |
| Com_show_new_master | 0 |
| Com_show_open_tables | 0 |
| Com_show_processlist | 119 |
| Com_show_slave_hosts | 0 |
| Com_show_slave_status | 0 |
| Com_show_status | 359 |
| Com_show_innodb_status | 0 |
| Com_show_tables | 174 |
| Com_show_variables | 45 |
| Com_slave_start | 0 |
| Com_slave_stop | 0 |
| Com_truncate | 0 |
| Com_unlock_tables | 0 |
| Com_update | 896216 |
| Connections | 544151 |
| Created_tmp_disk_tables | 2701 |
| Created_tmp_tables | 149449 |
| Created_tmp_files | 0 |
| Delayed_insert_threads | 0 |
| Delayed_writes | 0 |
| Delayed_errors | 0 |
| Flush_commands | 1 |
| Handler_commit | 0 |
| Handler_delete | 205829 |
| Handler_read_first | 795446 |
| Handler_read_key | 297198914 |
| Handler_read_next | 386510136 |
| Handler_read_prev | 5364182 |
| Handler_read_rnd | 28675629 |
| Handler_read_rnd_next | 338831974 |
| Handler_rollback | 0 |
| Handler_update | 1427521 |
| Handler_write | 7165838 |
| Key_blocks_used | 7764 |
| Key_read_requests | 749204576 |
| Key_reads | 4707203 |
| Key_write_requests | 1366298 |
| Key_writes | 1221891 |
| Max_used_connections | 70 |
| Not_flushed_key_blocks | 0 |
| Not_flushed_delayed_rows | 0 |
| Open_tables | 64 |
| Open_files | 112 |
| Open_streams | 0 |
| Opened_tables | 41237 |
| Questions | 8138106 |
| Qcache_queries_in_cache | 0 |
| Qcache_inserts | 0 |
| Qcache_hits | 0 |
| Qcache_lowmem_prunes | 0 |
| Qcache_not_cached | 0 |
| Qcache_free_memory | 0 |
| Qcache_free_blocks | 0 |
| Qcache_total_blocks | 0 |
| Rpl_status | NULL |
| Select_full_join | 82 |
| Select_full_range_join | 40622 |
| Select_range | 1829137 |
| Select_range_check | 0 |
| Select_scan | 988762 |
| Slave_open_temp_tables | 0 |
| Slave_running | OFF |
| Slow_launch_threads | 2 |
| Slow_queries | 222 |
| Sort_merge_passes | 0 |
| Sort_range | 753450 |
| Sort_rows | 104306761 |
| Sort_scan | 195086 |
| Table_locks_immediate | 11388636 |
| Table_locks_waited | 5749 |
| Threads_cached | 0 |
| Threads_created | 544150 |
| Threads_connected | 2 |
| Threads_running | 1 |
| Uptime | 114893996 |
+--------------------------+------------+
END

[AWS]

Is that a HT enabled P4?
The hardware you have should handle the load easily.
The main problem is using cgi instead of isapi. Switch to isapi and you'll notice the difference. Upgrade to 4.3.10 then install mmcache.
When installing php or mysql never use the installer package. Manually install both. Doing it this way makes upgrading much easier.
Post the contents of my.ini. If you can't open it because it is called my.cnf then rename it to my.ini.

[Dabbler]

Is that a HT enabled P4?No


The main problem is using cgi instead of isapi. Switch to isapi and you'll notice the difference.I think cgi/fcgi was used originally when we were on a NT2003 box and isapi didn't seem very stable. This "downgrade" to NT2000 just happened a few days ago and the config was kept the same. Just a choice made late at night by our "host" service.

[Dabbler]

Made the switch to ISAPI, and things appear to have improved considerably.

Forum loads are relatively light because of the weekend, so I may not get a chance to really stress test this until Monday.

[eva2000]

you should enable HT for 4 cpus :)

do a search on your server for a my.ini or my.cnf file should be at C:\my.ini and post it's contents

[Dabbler]

do a search on your server for a my.ini or my.cnf file should be at C:\my.ini and post it's contentsI don't have root access and I've literally asked my host service for this info half a dozen times in the last 2 weeks. I'm ready to go down there with a baseball bat.

[eva2000]

I don't have root access and I've literally asked my host service for this info half a dozen times in the last 2 weeks. I'm ready to go down there with a baseball bat.
hmm that doesn't sound good heh

[Dabbler]

Here is a cpu usage graph (Task Mgr) with 50-60 users online (5 min window).
This is a slightly bigger load than on last Friday with the switch to isapi.
Unfortunately, we still see cpu usage pinned at 100% for extended periods.
Page delivery times increase significantly at these load levels.

Next step, install mmcache.

No status yet on mysql config settings.

[Zachery]

From my understanding of things, the processor in windows even if pegged at 100% is just reserved most of the time, it may not truely be using all 100% of the avaible power but it might use it if it needs to.

[Dabbler]

I can only hope.

What I do experience is that as the count of "Currently active users" climbs, page response slows down. With 20-30 users, page delivery is normal. No noticeable delays. With 50-60 users, page delivery approaches 10 secs, sometimes going to 20-30 secs.

I'm not even near my peak loads, so I'm worried about the ability of the server to handle it.

Other minutiae:
1 - Attachments (~1 MB) are apparently in the data base. I read some comment that implied it's better if they were external to the database. How do I do that? Should I do that?

2 - Repair/Optimize functions time out (> 120 secs) for some of the larger tables (optimization of "post" comes to mind). Should I be concerned? What should I do?

[Dabbler]

Here's the ini file contents:

#This File was made using the WinMySQLAdmin 1.4 Tool
#7/19/2001 4:18:00 AM

#Uncomment or Add only the keys that you know how works.
#Read the MySQL Manual for instructions

[mysqld]
basedir=C:/mysql
#bind-address=206.173.31.52
datadir=C:/mysql/data
#language=C:/mysql/share/your language directory
#slow query log#=
#tmpdir#=
#port=3306
#set-variable=key_buffer=16M
[WinMySQLadmin]
Server=C:/mysql/bin/mysqld-nt.exe

______________________________________
Just got the above from my host service.
I guess this is just the default file.

[AWS]

Add this to my.ini and restart mysql.

skip-locking
set-variable = max_connections=500
set-variable = key_buffer=32M
set-variable = max_allowed_packet=16M
set-variable = sort_buffer=2M
set-variable = record_buffer=2M
set-variable = table_cache=1024
set-variable = thread_cache_size=164
set-variable = wait_timeout=1800
set-variable = connect_timeout=10
set-variable = max_connect_errors=10
set-variable = thread_concurrency=2
set-variable = myisam_sort_buffer_size=64M
query_cache_limit = 2M
query_cache_size = 32M
query_cache_type = 1
[mysqldump]
quick
set-variable = max_allowed_packet=32M

[mysql]
no-auto-rehash

[isamchk]
set-variable = key_buffer=256M
set-variable = sort_buffer=256M
set-variable = read_buffer=2M
set-variable = write_buffer=2M

[myisamchk]
set-variable = key_buffer=256M
set-variable = sort_buffer=256M
set-variable = read_buffer=2M
set-variable = write_buffer=2M

Also get mmcache installed.
You'll notice the difference right away once you add the above to my.ini and restart mysql.

[Dabbler]

Many thanks.

Hopefully it won't take as long to get them to upload this new file as it took to download the old.

Stay tuned...

[Dabbler]

Attached is a cpu utilization graph (Task Mgr) with the new my.ini installed (but no mmcache yet - "installed but not being called" per the host).

So, with moderate loads (80-90 users - 5 minute window), the server seems to be handling it adequately, just occasional moments of the cpu pinned at 100%.

__________________________________
Once mmcache is made to work, I expect another slight improvement. But I don't think this platform will allow me to hit my target of 300 concurrent users (5 min win). So, what do I need?

[AWS]

You'll notice a marked improvement with mmcache.
Windows will peak at 100% cpu load. Don't read too much into it. If it hits 100% and stays there then you have a problem. If that happens it means you have a run away script.

[Dabbler]

Well, problems continue.

Although there was a performance improvement with the new file,
the server experienced lock ups with even a moderate load.
After 4 reboots (and apparently some table problems) we just
downgraded to the original my.ini file. Everything seems much more
stable.

Also, the 2 techs working on this can't get mmcache to run.
They are looking at eaccelerator (sp?).

This is just crazy.

How about I just get a machine at theplanet? I can have the
site transferred there. What is the protocol for having some
local talent from here configure this thing properly for me?
What kind of hardware do I need? I am so tired of wasting
my time on this!

[AWS]

Sounds like you host doesn't have much experience with Windows servers. MMcache is very easy to install on Windows.
What version of mmcahe are you using?
You are running php 4.3.8 and if you are trying to use the mmcache zip I added to another post here then it won't work. I compiled it for php 4.3.10 and it will only work with that version of php. I am not even sure the source is available for older versions of php any longer.
I've been running Windows servers exclusively for almost 4 years now and I have never had problems with server performance. I think maybe you should look into getting a new host. I guarantee that once you get a Windows server setup correctly you'll be happy.

[Dabbler]

MMcache is very easy to install on Windows.
What version of mmcahe are you using? You are running php 4.3.8 and if you are trying to use the mmcache zip I added to another post here then it won't work. I compiled it for php 4.3.10 and it will only work with that version of php.They installed php 4.3.10 last night. That went smoothly. They now understand the mmcache version problem (hence the upgrade to 4.3.10) but the tech said that he still couldn't get mmcache to run. He said he was using the one compiled for 4.3.10, so presumably he used the right version. It's just install a dll, call it from the ini file and reboot? Do they need to recompile for some reason?


I've been running Windows servers exclusively for almost 4 years now and I have never had problems with server performance. I think maybe you should look into getting a new host. I guarantee that once you get a Windows server setup correctly you'll be happy.As mentioned previously, I'm keen to get spiderable archives, which I understand only works under apache.


Sounds like you host doesn't have much experience with Windows servers.It's their "specialty." :rolleyes:

[Dabbler]

Open Issues:

1 - Attachments (~1 MB) are apparently in the data base. I read some comment that implied it's better if they were external to the database. How do I do that? Should I do that?

2 - Repair/Optimize functions time out (> 120 secs) for some of the larger tables (optimization of "post" comes to mind). Is this RAM intensive? Data + Index > 1GB, so is 1 GB of RAM sufficient? Should I be concerned? What should I do?

3 - The my.ini file was downgraded yesterday morning to the vanilla default file because of stability problems. Should last night's the php upgrade positively affect this?

4 - mmcache not working.

5 - Still need recommendations for an appropriate platform.

6 - Can a dual processor system be P4-based or only Xeon-based? Intel doesn't (currently) list any dual processor capable P4 cpu's. I ask because my host can't tell me if the dual cpu machine I'm currently on is a Xeon or not. They think it is only a P4.

7 - Rather than continue to wrestle with this host, how about I just get a machine at theplanet? I can have the site transferred there. What is the protocol for having some local talent from here configure this thing properly for me?

[AWS]

There are no dual capable P4s except Xeons. You might have dual Xeons. Windows 2000 will not see the virtual processors of Xeons. Win2K is not HT aware. The only way Win2K will see all 4 is if the host uses a 4 processor license. Makes me wonder what kind of host they are if they don't even know what kind of server they have. If you want to know what type of processor you have connect with remote desktop and right click my computer and select properties. That will tell you what processor you have.
Installing mmcahe is pretty easy. Add a few lines to php.ini and restart IIS. They probably have the path to mmcache set wrong or are not using the Zend extension. On Windows you have to use the Zend extension.
A couple things after looking at your phpinfo.
Path to php.ini is set to C:\WINNT. Is php.ini in that dir? If it was it would show C:\WINNT\php.ini
extension_dir is set to C:\php4. Are your extensions actually in that dir. They are usually found in C:\php4\extensions unless your host moved then to php root.

[Dabbler]

There are no dual capable P4s except Xeons. You might have dual Xeons. Windows 2000 will not see the virtual processors of Xeons. Win2K is not HT aware. The only way Win2K will see all 4 is if the host uses a 4 processor license. Makes me wonder what kind of host they are if they don't even know what kind of server they have.I think I may have answered that in Post 10 above. :rolleyes:

Anyway, the techs insist that it is a P4, no HT, and not a Xeon. Early this week I asked for the part number off the cpu housing, but they weren't at the CO to look. He offered to take a picture of it for me, but I haven't seen anything yet.


If you want to know what type of processor you have connect with remote desktop and right click my computer and select properties. That will tell you what processor you have.I don't have an admin account on this machine, nor do I have root access. Is it still possible to access remotely? How would I do that?


Installing mmcache is pretty easy. . .A couple things after looking at your phpinfo. . . Thanks for the suggestions!!

[Dabbler]

This cpu performance graph (TaskMgr) shows the load with about 70 users (5 min win). Near the end the load is pinned to 100%. A few minutes later page delivery slowed to over 30 seconds, basic functions started timing out, data base error messages were reported to visitors, and connections to the server were being refused. Then it got bad.

[Dabbler]

On a positive note, the tech handling the mmcache install (their hw guru) indicated that php.ini was in c:\php because that is were the install instructions recommended it be placed. (Thanks AWS for the tip!) He wants to move it to the c:\winnt directory, perhaps tonight because we need to reboot to test any mmcache install.

[AWS]

I don't have an admin account on this machine, nor do I have root access. Is it still possible to access remotely? How would I do that?
Login to RDP with your user account that you log in to ftp or shell with.
Looks like you got mmcache working. Is it making things better?
On the mysql problems. Is there more than one site on the box using mysql?

[Dabbler]

A little progress.

The tech successfully installed mmcache (yea).
Zend Engine v1.3.0, Copyright (c) 1998-2004 Zend Technologies with Turck MMCache v2.4.7-CVS, Copyright (c) 2002-2003 TurckSoft, St. Petersburg, by Dmitry Stogov
After about 12 hours of activity, we have:
Caching Enabled true
Optimizer Enabled true
Memory Size 16,777,176 Bytes
Memory Available 3,759,120 Bytes
Memory Allocated 13,018,056 Bytes
Cached Scripts 97
Removed Scripts 0
Cached Keys 0

User activity is light because of the weekend. I won't be able to stress the system until Mon/Tues when user activity will grow by 2X-3X.

Questions:


Since I am using php 4.3.10, do I need to worry about the serialize bug? I couldn't find any focused discussion of the issue here on these boards of what it is and how it manifests itself. (Only found tangential asides in unrelated threads.)
Is the default config of mmcache sufficient?
I still don't have an optimized my.ini file (still using the default). Is there any reason why the revised ini would have caused problems (see my Post 20 above)? Do I need a newly revised ini? Or should I try to use the same revised ini file again?
What else do I need to provide to get a hw recommendation?

[Dabbler]

Login to RDP with your user account that you log in to ftp or shell with. RDP?? :confused:
I do have ftp access to a user (non-root) account.

On the mysql problems. Is there more than one site on the box using mysql?They tell me that we are the sole site on a dedicated machine.

[AWS]

Try the my.ini from my post above. I changed it a bit. Also make sure you load mysqld-max-nt.exe.
The settings you have for mmcache look good.

[Dabbler]

This'll be my.ini:

[mysqld]
basedir=C:/mysql
datadir=C:/mysql/data
skip-locking
set-variable = max_connections=500
set-variable = key_buffer=32M
set-variable = max_allowed_packet=16M
set-variable = sort_buffer=2M
set-variable = record_buffer=2M
set-variable = table_cache=1024
set-variable = thread_cache_size=164
set-variable = wait_timeout=1800
set-variable = connect_timeout=10
set-variable = max_connect_errors=10
set-variable = thread_concurrency=2
set-variable = myisam_sort_buffer_size=64M
query_cache_limit = 2M
query_cache_size = 32M
query_cache_type = 1

[mysqldump]
quick
set-variable = max_allowed_packet=32M

[mysql]
no-auto-rehash

[isamchk]
set-variable = key_buffer=256M
set-variable = sort_buffer=256M
set-variable = read_buffer=2M
set-variable = write_buffer=2M

[myisamchk]
set-variable = key_buffer=256M
set-variable = sort_buffer=256M
set-variable = read_buffer=2M
set-variable = write_buffer=2M

[WinMySQLadmin]
Server=C:/mysql/bin/mysqld-nt.exe

[Dabbler]

Here's part of the problem:

CPU1: X86 Family 6 Model 8 Stepping 6 Genuine Intel ~1004 MHz
CPU2: X86 Family 6 Model 8 Stepping 6 Genuine Intel ~1004 MHz

HDD: Mylex AcceleRAID 170 SCSI Disk (67 GB formatted)
DC1: Mylex AcceleRAID 170 Disk Array Controller
DC2: Mylex AcceleRAID 170 Disk Array Controller (error)

I'm not exactly sure, but my best guess is that the CPU's are 1 GHz P3 Coppermines. There are also Celerons and Xeons in this named family (6-8-6).(Thanks Intel for clarity in your naming conventions. :rolleyes: ).

Relatively slow by today's standards.

[Zachery]

If you have 2, they are xeons.

[AWS]

You have dual P3 1GHz processors and they more than likely are Coppermines unless they are older slot1 cpu. Celerons won't run as duals. The Mylex 170 is more than likely the cause of the problems. I have the same controller and when used with raid 5 it causes mysql to run slow at times. Make sure you have the cache on the controller set to on. This will speed up the reads and writes.
That server should handle the load. Dual P3s are workhorses although they aren't used any longer by many hosting companies. Most hosts opt for HT enabled P4s or Xeons in place of them and for economy hosting Celerons.

[Zachery]

I wasn't aware you could do dual P3's :) my bad.

[AWS]

I wasn't aware you could do dual P3's :) my bad.
I still have 3 of them. One is a dev box and the other 2 are for sale.

[Dabbler]

The Mylex 170 is more than likely the cause of the problems. I have the same controller and when used with raid 5 it causes mysql to run slow at times. Make sure you have the cache on the controller set to on. This will speed up the reads and writes.Where is that done exactly?

That server should handle the load.Really? I hope you're right. It's struggling at a fraction of the load I need to handle.

Two steps left then:
1 - replace my.ini today.
2 - check mylex controller config

[AWS]

There is a program available at Mylex site that allows you to configure the controller. You will need to use this since you don't have physical access to the server. You have to do it when the bois loads or use the tools in Windows.
I didn't move off dual P3 until I started to get 900+ online all the time. I have since sold that site and my current sites are on HT enabled P4.

[Dabbler]

There is a program available at Mylex site that allows you to configure the controller. You will need to use this since you don't have physical access to the server. You have to do it when the bios loads or use the tools in Windows.Thank you for your continuing kindness and support.

I didn't move off dual P3 until I started to get 900+ online all the time. I have since sold that site and my current sites are on HT enabled P4.Do you recall the cookie timeout you were using?

[AWS]

900 seconds.

[Dabbler]

I won't be able to stress the system until Mon/Tues when user activity will grow by 2X-3X.This morning, with ~65 users (5 min win), the CPU's were literally pinned to 100% for the duration of the TaskMgr session. Page delivery was very slow (>20 secs). Very bad performance on all tasks. Yuck.

I'll post the TaskMgr snap when I get a chance.

We will be upgrading my.ini tonight.

[Dabbler]

Here is a Task Manager screen from this morning.

Users: ~70 (5 min win)

CPU's are pinned to 100%. Forums quite slow.

The pinning is intermittent. That is, CPU utilization fluctuates between 20 and 99% from moment to moment, and then all of a sudden goes straight to 100% and stays there for 1-5 minutes. No unusual forum activity that I can spot. Then the machine recovers and CPU usage fluctuates again.

Forum recovered nicely this time. No db errors reported (that I am aware of.)
But I am still far from my target server load.

[AWS]

When that happens look at under the processes tab and you should be able to find what is using up the most resources. Second column shows cpu usage. It will be easy to spot the culprit. Take a screenshot of that screen and post it when you hit the 100% mark.

[Dabbler]

Not going to happen because they've locked me out of RDP.

So, back to my original question:
What hw do I need for 600-750 users (900 min win) given the above?

[AWS]

Ideally you should go with 2 servers although your current server should hold up under that load. One for http and one for mysql. You could use your existing server for http and add a dual Xeon with at least 2 gigs ram for the db server. This would future proof the site also. You wouldn't have to worry about an upgrade for a long time.
I wish you were able to post the processes during the slowdown and cpu spike.
Something is hogging the resources and I don't think it has anything to do with mysql or vbulletin.

[Dabbler]

Ideally you should go with 2 servers although your current server should hold up under that load. One for http and one for mysql. You could use your existing server for http and add a dual Xeon with at least 2 gigs ram for the db server. This would future proof the site also. You wouldn't have to worry about an upgrade for a long time.A couple points:

1 - I apologize for being insufficiently clear in my initial posts, but the current platform is strictly temporary. I was on a P4 2.8 that was severely hamstrung. Although we paid thousands to set it up and thousands to "fix" it, it lasted only 9 months and crashed hard at the end of Feb (in no small part because their server management/monitoring tools were MIA). Our host service moved us on to this temp platform which they claimed had 2x P4 2.4 GHz processors, double the RAM and faster hard drives. Cool, I thought, this should hum. But performance did not improve. And now I know why. (Thanks to all who helped, especially AWS!)

2 - I lean more toward a fast idiot-proof setup versus one that needs to be tweaked alot to achieve equivalent speed. Not that it wouldn't be fun, but I don't have time to babysit the hw in order to make sure it is working. Nor do I have days on end to play with config files to see if I can make the system to run a little faster. (NOTE: the newly tweaked my.ini file was replaced with the original my.ini file again this morning after the forums crashed. I am happy to take it on faith that an optimized ini file will boost performance. But this is just a temp server and I'm well past the point of diminishing returns.)

3 - Put the db on a dual xeon or a dual opteron? I've seen many comments recommending AMD for the db. How difficult is it to set up a 2 server solution? How delicate (or robust) is a 2 server solution? (I'm a little gun shy at the moment. :o ) Would an opteron make a good 1 box solution?


I wish you were able to post the processes during the slowdown and cpu spike. Something is hogging the resources and I don't think it has anything to do with mysql or vbulletin.I wish I was able to as well. But someone needs to monitor site traffic and server loads real-time. They won't and they won't let me. But that's another thread...

[AWS]

If you go dual opteron or dual xeon you can't go wrong. I lean to Intel because it is the platform I know the best. If you go with a 2 server setup you won't have to worry about it once it is setup correctly for a long time, more than likely never.
If you go with a good host they will set the servers up correctly as long as you tell them what you want to do. Once the 2 servers are setup then hire someone to configure the servers and move the sites.
If you want to go with one box then get either a dual opteron or dual xeon. Make sure you have at least 2 gigs ram installed. Make sure it has 2 hard drives. One will be used for mysql and the other for the OS and webroot. Make sure that if you go raid don't use raid 5. In fact stay away from raid and go with 2 scsi drives or 2 sata drives. Once you decide on the server and after the host puts it online hire someone to set it all up.
When looking for a host, I would not stay with your current host, talk to them and ask them what they would reccomend. Make sure you get root access to the box. I don't know why your host doesn't give you root access to your deicated server unless it really isn't dedicated.

[Dabbler]

Make sure it has 2 hard drives. One will be used for mysql and the other for the OS and webroot. Make sure that if you go raid don't use raid 5. In fact stay away from raid and go with 2 scsi drives or 2 sata drives.So many here seem to recommend RAID. No performance boost? Or too many headaches?

[AWS]

Raid is great. If you want raid use raid 10 instead of raid 5. For me the performance increase I saw when I used raid just wasn't worth the extra cost. A single fast scsi or sata drive performs just about as well.

[Dabbler]

The Mylex 170 is more than likely the cause of the problems. I have the same controller and when used with raid 5 it causes mysql to run slow at times. Make sure you have the cache on the controller set to on. This will speed up the reads and writes.Here's the Mylex configuration from this server:

----
Name: Logical Drive 0
Status: On
Capacity: 67.797 Gb
RAID: 5
Cache Line Size: 8 Kb
Stripe Block Size: 64 Kb
Write Cache: Write-Thru
----


Separately, the Mylex AcceleRAID 170 spec sheet:
http://www.bellmicro.com/fibrechannel/newasp/Mylex/Downloads/AcceleRAID170.pdf
indicates:
Ensure maximum data protection with ECC cache, supporting 32MB or 64MB DIMM memory.
and
Flash 1MB
Cache 32MB or 64MB

[Dabbler]

Last bump?