Hypography Science Forums: alexander - Viewing Profile - Hypography Science Forums

So yeah... best network topology diagram i have produced yet... as you can see it's for a load balancer...
 IMG_20110809_095316.jpg (87.27K)
Number of downloads: 75

So recently, i've been toying a bit with MySQL, and it is amazing how a proper setup, and a proper server configuration can affect the performance of MySQL. So to give some ideas about how i am going about and setting this up and how it affected throughput, here is a brief tear-down.

Server: Every time its a VM, 4 physical cores (no ht atm), 8GB ram, 80GB hard drive capacity.
Setup 1 (one i'm trying to show is not very optimal):
RHEL 5.6 stock
MySQL 5.0 (latest release, whatever it is, 92 i think) installed from rpm
originally stock config

Setup 2 (one i'm working with):
CentOS 5.6
3 partitions, boot (ext3), / (ext3) and /var/lib/mysql (XFS)
/var/lib/mysql is mounted with noexec,noatime,nodiratime,nobarrier
vm.swappiness = 0
vm.dirty* = 0
MySQL 5.5 (rpm, same deal)
google perf tools
tcmalloc_minimal preloaded pre mysqld start
not-so-stock config includes:

# InnoDB options
#Do not extend files too frequently
innodb_autoextend_increment = 20
# Use one file per table
innodb_file_per_table
# Buffer pool size
innodb_buffer_pool_size           = 5G
innodb_additional_mem_pool_size   = 32M
# Transaction log location and sizes
innodb_log_files_in_group         = 4
innodb_log_file_size             = 256M 
innodb_log_buffer_size            = 4M
# Percentage of unwritten dirty pages not to exceed
innodb_max_dirty_pages_pct        = 80
# Transaction commit policy
innodb_flush_log_at_trx_commit    = 2
# Timeout to wait for a lock before rolling back a transaction
innodb_lock_wait_timeout          = 50
# Flush method
innodb_flush_method               = O_DIRECT
# Number of concurrent threads to run
innodb_thread_concurrency         = 8
# Autoinc lock mode
innodb_autoinc_lock_mode = 1
# Enable fast innodb shutdown
innodb_fast_shutdown              = 0
# Dont delay insert, update and delete operations when purge
# operations are lagging
innodb_max_purge_lag              = 0

So, what did i try.
I used mysqlslap to test mysql io performance, so i ran it with a concurrency of 1, for 100000 queries write-centric, the results of the first run were vastly different. It took over 40 sec on the stock setup, so i fixed that by changing the setup to resemble what i am using in 5.5, here are the results:
Example of command line: mysqlslap --concurrency=1 --iterations=1 --engine=innodb --auto-generate-sql --auto-generate-sql-load-type=write --number-of-queries=100000 -u [user] -h [ip] --password=[password]
stock:

Quote

mine:

Quote

So i increased concurency to see how they bare:
stock:

Quote

it doesn't crash, but queries start timing out, so you get a "server unresponsive thing here" the mysql server doesn't actually crash

not-so-stock:

Quote

no conclusions yet, though you can see, that server setup and configuration does matter and can gain significant performance on the same hardware. Now, i'm going to write a script with a battery of tests and attempt to brake both setups, and we'll see how both of these servers bare the weight, write though, as you can see, there is a significant difference in performance already I also want to monitor io, threads, and stuff like that while its all running...

For now, this was the first part of a teaser...

So if i haven't been active, it's been in part because i have been readying something for a first release. It's an open-source light-weight PHP framework. It's based on a reflection API, and it does some magic stuff, like it will automatically generate a WSDL (SOAP) and JSON APIs to your code, using your comments for documentation. It's neat, it's light and it will be beta released shortly (within a couple of weeks).

If you are interested, and have time, i have some neat ideas about the framework, and it would be nice to see other people using it (though atm i dont have available examples of how to use it, i haven't written new/open ones, but i can explain; it's because the code i wrote for it (lots and lots of it) is proprietary (product of a company i used to work for))

Anyways, i'll explain further if there is any interest...

So we have had a similar discussion a while back titled "you know you are a geek when", but how do you know for sure if you are talking to a geek.

I'll start off:

You know you are talking to a geek when they have a nerdgasm when you tell them that all you are trying to do is setup an ipsec tunnel over mpls on an ospf network between two nssa asbr routers such that this traffic falls into your cs5 qos domain in the 4th queue at 3% max...

You know you are talking to a geek when the term "double D" is met with a perplexing look pondering how one can have a two-sided die...

You know you are talking to a geek when the term "trial and terror" is both endearing and relaxing...

You know you are talking to a geek when they solve a tedious data entry problem with a 15 line script and threaten to replace you with it...

You know you are talking to a geek when the conversation starts with "So i was fuzzing against [insert any application here] last night"...

So wondering what kind of a relationship you have with your company's IT. If you are IT, i know how you feel about people, but if you have other nerdy groups in your organization, how do you feel about their requests for access to stuff you manage? Do they get access?