Re: RAM-usage and hardware upgrade 10gb RAM

2004-07-23 Thread Jan Kirchhoff
David Griffiths wrote:
We just put a new dual-Opteron server into our production environment. 
We ordered a Megaraid SCSI card and five 10k drives, and a 3Ware 
Escalade SATA card with six 7200 RPM drives (Maxtor) to see which ones 
were best.

Our network guy did a bunch of benchmarking on the drives and found 
that SCSI-RAID5 was a bit faster than SATA-RAID0+1.

The SATA was significantly cheaper (the 3Ware card was the same price 
as the Megaraid card, however). You might be able to tie a 10K SCSI 
rig if you went with the Western Digital Raptor drives.

We ended up putting the SATA drives in production - some bug in the 
SCSI driver kept crashing MySQL on index-creation, etc.

High Performance MySQL mentions that SCSI 15K drives are worth the 
extra money.

Thanks David for your post,
Does anybody else in this list have experience with SATA-RAIDs?
After having done some research it looks like we'll go with a 
dual-Opteron an 8-12GB of RAM and a SATA-RAID10 with 8-10 
250GB-SATA-discs. We are just waiting for the NCQ-SATA-drives to be 
available and for 2 colleagues to return from vacation since we want 
everybody to be here when we do that major change. (looks like we'll 
order the system in 2-3 weeks if the harddiscs are available)

Our most important tables that get selects all the time and get updated 
up to 30 times a second each (or even more often depending on the time 
of the day) are of a total size of about 5-6 gigs.
Is it realistic thinking that mysql/innodb would keep those tables 
totally in memory and reply to all selects without reading from the disc 
when we increase innodb_buffer_pool_size to 7 or 8 gigs (assuming we 
have 12gigs of RAM)?

I just wanted to make sure nobody has hit problems with such systems. If 
you could just send a short We're doing something like that and it 
works fine I could definitly sleep better ;)

thanks for all the posts so far and pointing me towards the right direction!
Jan
--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:http://lists.mysql.com/[EMAIL PROTECTED]


Re: RAM-usage and hardware upgrade 10gb RAM

2004-07-20 Thread Egor Egorov
Jan Kirchhoff [EMAIL PROTECTED] wrote:

 Money is not really an issue but of course we don't want to waste it for 
 scsi-hardware if we can reach almost the same speed with hardware 
 sata-raids.

'Almost' is a key word. Some SCSI disk are working at 15k RPM, which will give
you a HUGE MySQL performance growth compared to 10k disks. 

AFAIR, there are no 15k RPM SATA disks yet. 


 Money is not really an issue but of course we don't want to waste it for 
 We'd like to stay with x86 because all our hardware is intel/amd and all 
 our servers are running debian-linux. Can we expect better performance 
 or problems using kernel 2.6.x?

You can expect better performance on kernel 2.6.x of course, especially on
multiple requests. 






-- 
For technical support contracts, goto https://order.mysql.com/?ref=ensita
This email is sponsored by Ensita.net http://www.ensita.net/
   __  ___ ___   __
  /  |/  /_ __/ __/ __ \/ /Egor Egorov
 / /|_/ / // /\ \/ /_/ / /__   [EMAIL PROTECTED]
/_/  /_/\_, /___/\___\_\___/   MySQL AB / Ensita.net
   ___/   www.mysql.com




-- 
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:http://lists.mysql.com/[EMAIL PROTECTED]



Re: RAM-usage and hardware upgrade 10gb RAM

2004-07-20 Thread Jan Kirchhoff
Egor Egorov wrote:
Money is not really an issue but of course we don't want to waste it for 
scsi-hardware if we can reach almost the same speed with hardware 
sata-raids.
   

'Almost' is a key word. Some SCSI disk are working at 15k RPM, which will give
you a HUGE MySQL performance growth compared to 10k disks. 

AFAIR, there are no 15k RPM SATA disks yet. 
 

But shouldn't a sata-based RAID10 with 8 discs do job as well? writes 
would be spread on 4 discs...
Has anybody experience with those external SCSI-to-SATA RAIDs?
A SCSI-solution would cost twice as much, but would it really speed 
things up compared to a massive use of parallel (raid0) sata-discs?
I know disc i/o is the bottleneck in our case, of course we want the 
fastest disc/raid-system we can possibly get for our money.
Is our thinking too simple or shouldn't it be possible to reach the 
speed of fast scsi-discs by simply taking 2-3 fast sata-discs in a 
hardware raid0?
Our goal is a raid10, so reading should be even faster.

Money is not really an issue but of course we don't want to waste it for 
We'd like to stay with x86 because all our hardware is intel/amd and all 
our servers are running debian-linux. Can we expect better performance 
or problems using kernel 2.6.x?
   

You can expect better performance on kernel 2.6.x of course, especially on
multiple requests. 
 

Has anybody experiences with RAM-usage and cpu-architecture (please have 
a look at my earlier post)?

thanks
Jan
--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:http://lists.mysql.com/[EMAIL PROTECTED]


Re: RAM-usage and hardware upgrade 10gb RAM

2004-07-20 Thread Brent Baisley
The single biggest difference between SATA (or any IDE) and SCSI is 
command queuing. Command queuing allows the drive to intelligently 
reorder reads and writes to make things faster.
An ATA drive executes a bunch of commands in the order it gets them, 
which can be slow if it needs to write data on an inside track, then 
read from and outside track, and back and forth until the both the read 
and write requests are finished.
SCSI will understand that there is a proximity benefit to the commands, 
so it will reorder the interlaced requests and execute one before the 
other. Essentially putting part of one request on hold while it does 
the other. That's an oversimplification of the algorithm of course. 
Especially since a server system will probably have many more the two 
interlaced requests.

Did you ever try to clean two rooms at once? Put the clothes away in 
one, then the other. Make the bed in one, then the other. Lots of 
useless travel time in there. You would probably clean one and then the 
other. Except if you are vacuuming, then you would probably vacuum both 
at once, interlacing two similar actions.

SATA is catching up though. Seagate has release SATA drives that have 
command queueing, but I don't know how it compares to SCSI.

On Jul 20, 2004, at 10:40 AM, Jan Kirchhoff wrote:
But shouldn't a sata-based RAID10 with 8 discs do job as well? writes 
would be spread on 4 discs...
Has anybody experience with those external SCSI-to-SATA RAIDs?
A SCSI-solution would cost twice as much, but would it really speed 
things up compared to a massive use of parallel (raid0) sata-discs?
I know disc i/o is the bottleneck in our case, of course we want the 
fastest disc/raid-system we can possibly get for our money.
Is our thinking too simple or shouldn't it be possible to reach the 
speed of fast scsi-discs by simply taking 2-3 fast sata-discs in a 
hardware raid0?
Our goal is a raid10, so reading should be even faster.

--
Brent Baisley
Systems Architect
Landover Associates, Inc.
Search  Advisory Services for Advanced Technology Environments
p: 212.759.6400/800.759.0577
--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:http://lists.mysql.com/[EMAIL PROTECTED]


Re: RAM-usage and hardware upgrade 10gb RAM

2004-07-20 Thread David Griffiths
We just put a new dual-Opteron server into our production environment. 
We ordered a Megaraid SCSI card and five 10k drives, and a 3Ware 
Escalade SATA card with six 7200 RPM drives (Maxtor) to see which ones 
were best.

Our network guy did a bunch of benchmarking on the drives and found that 
SCSI-RAID5 was a bit faster than SATA-RAID0+1.

The SATA was significantly cheaper (the 3Ware card was the same price as 
the Megaraid card, however). You might be able to tie a 10K SCSI rig if 
you went with the Western Digital Raptor drives.

We ended up putting the SATA drives in production - some bug in the SCSI 
driver kept crashing MySQL on index-creation, etc.

High Performance MySQL mentions that SCSI 15K drives are worth the extra 
money.

Fast hard drives are important, but so is lots of RAM (which is where 
the Opteron shines). In fact, all the benchmarks I've seen show that the 
Opteron/Athlon architecture beats Intel processors by a 30-odd percent 
margin if memory serves (note that for some reason, most benchmarks I've 
seen were on 3.23, which is outdated and not overly usefull).

One of our websites serves up 2 million distinct pages per day; the 
original coders of the site did something dumb and open a new connection 
to the database for most of those pages (probably about 1.8 million). 
Even with that additonal load, our Opteron server has an average CPU 
load of about 10%.

David
Jan Kirchhoff wrote:
Egor Egorov wrote:
Money is not really an issue but of course we don't want to waste it 
for scsi-hardware if we can reach almost the same speed with 
hardware sata-raids.
  

'Almost' is a key word. Some SCSI disk are working at 15k RPM, which 
will give
you a HUGE MySQL performance growth compared to 10k disks.
AFAIR, there are no 15k RPM SATA disks yet.  

But shouldn't a sata-based RAID10 with 8 discs do job as well? writes 
would be spread on 4 discs...
Has anybody experience with those external SCSI-to-SATA RAIDs?
A SCSI-solution would cost twice as much, but would it really speed 
things up compared to a massive use of parallel (raid0) sata-discs?
I know disc i/o is the bottleneck in our case, of course we want the 
fastest disc/raid-system we can possibly get for our money.
Is our thinking too simple or shouldn't it be possible to reach the 
speed of fast scsi-discs by simply taking 2-3 fast sata-discs in a 
hardware raid0?
Our goal is a raid10, so reading should be even faster.

Money is not really an issue but of course we don't want to waste it 
for We'd like to stay with x86 because all our hardware is intel/amd 
and all our servers are running debian-linux. Can we expect better 
performance or problems using kernel 2.6.x?
  

You can expect better performance on kernel 2.6.x of course, 
especially on
multiple requests.  

Has anybody experiences with RAM-usage and cpu-architecture (please 
have a look at my earlier post)?

thanks
Jan


--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:http://lists.mysql.com/[EMAIL PROTECTED]


Re: RAM-usage and hardware upgrade 10gb RAM

2004-07-20 Thread Marc Slemko
On Mon, 19 Jul 2004 18:13:36 +0200, Jan Kirchhoff [EMAIL PROTECTED] wrote:
 Hi,
 
 We are currently using a 4.0.16-replication-setup (debian-linux, kernel
 2.4.21, xfs) of two 2.4ghz Intel-Pentium4 systems with  3gig RAM each
 and SCSI-Hardware-Raid, connected via gigabit-ethernet. We are reaching
 the limit of those systems and are going to buy new hardware as well as
 upgrade to mysql 4.1.x. We will start testing our applications on 4.1.3
 within the next few weeks but our main problem is that we are not quite
 sure what hardware to buy...
 
 We are planning to buy something like a dual-xeon system with 10-16gb of
 RAM and hardware raid10 with 8 sata-disks and as much cache as possible.
 Will mysql be able to use the ram efficiently or are we hitting limits?
 AMD or Intel? 32bit or 64bit?

Whatever you do, get a 64 bit system.  Opteron recommended, if you
really prefer Intel and can get your hands on one of their 64-bit
Xeons that is acceptable, although it may take a little longer for
Linux to catch up.

Even if the software isn't there yet (it is, it may just be a bit of a
hassle to all get working), in the worst case you'll have to run it in
32-bit mode until you can figure that out.

You can't use more than 2 gig most of the time / close to 4 gig if you
hack things up right innodb cache on a 32 bit system.  The rest of the
memory will be used by the OS (less efficiently than on a 64-bit
system though), but that may or may not be as efficient as innodb
doing it.  That depends a lot on your application's data access
patterns.

-- 
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:http://lists.mysql.com/[EMAIL PROTECTED]



RAM-usage and hardware upgrade 10gb RAM

2004-07-19 Thread Jan Kirchhoff
Hi,
We are currently using a 4.0.16-replication-setup (debian-linux, kernel 
2.4.21, xfs) of two 2.4ghz Intel-Pentium4 systems with  3gig RAM each 
and SCSI-Hardware-Raid, connected via gigabit-ethernet. We are reaching 
the limit of those systems and are going to buy new hardware as well as 
upgrade to mysql 4.1.x. We will start testing our applications on 4.1.3 
within the next few weeks but our main problem is that we are not quite 
sure what hardware to buy...

We are planning to buy something like a dual-xeon system with 10-16gb of 
RAM and hardware raid10 with 8 sata-disks and as much cache as possible.
Will mysql be able to use the ram efficiently or are we hitting limits? 
AMD or Intel? 32bit or 64bit?
Money is not really an issue but of course we don't want to waste it for 
scsi-hardware if we can reach almost the same speed with hardware 
sata-raids.
We'd like to stay with x86 because all our hardware is intel/amd and all 
our servers are running debian-linux. Can we expect better performance 
or problems using kernel 2.6.x?
If it really adds performance we might change to something else but x86 
or change the OS, but definitly not for 2-5%.
We are going to keep the old servers as replication-slaves for big, time 
consuming selects and making backups.

We will have around 60,000,000 inserts/updates a day and lots of selects 
with joins on tables of all sizes (historical tables with 400,000,000 
rows as well as small tables with less than 500,000 rows) The whole size 
of the database will be around 200gb, growing up to 400gb in the next 12 
months. We are using innodb because we had big problems with the 
locking-issues of myisam. Some of the smaller tables that are updated 
all the time can be kept in memory if possible since its data is also 
cached/backuped by the applications that insert/update the data. Has 
anybody experienced problems with a innodb_buffer_pool_size 10gb?
Disk-I/O is our main problem since all the updates go to various tables 
spread on the discs. Since most of the data can be reconstruted in case 
of a crash it is ok for us to have delayed inserts and inserts being 
cached in memory. Are there more options for innodb-tables than 
increasing innodb_buffer_pool_size and setting 
innodb_flush_log_at_trx_commit=0 that could speed up inserts/updates?

thanks for any help/suggestions...
Jan
--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:http://lists.mysql.com/[EMAIL PROTECTED]