Re: DataMemory and IndexMemory - Mikael Ronström - com.mysql.lists.cluster

4 messages in com.mysql.lists.clusterRe: DataMemory and IndexMemory

From	Sent On	Attachments
Jim Hoadley	11 Apr 2005 12:29
Jim Hoadley	12 Apr 2005 11:05
Mikael Ronström	12 Apr 2005 13:48
Stewart Smith	12 Apr 2005 19:45

Subject:	Re: DataMemory and IndexMemory
From:	Mikael Ronström (mik...@mysql.com)
Date:	04/12/2005 01:48:48 PM
List:	com.mysql.lists.cluster

Hi Jim,

2005-04-12 kl. 20.05 skrev Jim Hoadley:

A related question:

When I increase DataMemory from 2600M to 2700M, my NDB nodes won't restart. Error given is:

Date/Time: Tuesday 12 April 2005 - 10:50:25 Type of error: error Message: Memory allocation failure Fault ID: 2327 Problem data: DBTUP could not allocate memory for Page Object of reference: Requested: 32768x86400 = 2831155200 bytes ProgramName: ndbd ProcessID: 15489 TraceFile: /usr/local/mysql/ndb_1_trace.log.12 ***EOM***

There is no entry in the management node error log.

Now I'm increasing (DataMemory + IndexMemory) from 3.1GB to 3.2GB, so it could an operating system issue, but since I'm running the hugemem kernel, a single process should be able to allocate up to 4GB of RAM.

It is an OS issue, the OS can't allocate the memory needed, there is also other memory needed in the process for general buffers, transaction records, operation records (probably around 300-400 MByte for your configuration.

Hmm, is there some other parameter I need to change? Such as NoOfDiskPagesToDiskAfterRestartTUP or NoOfDiskPagesToDiskAfterRestartACC?

You can increase them if you like, that will speed up system restarts somewhat. During system restart one doesn't need the same carefulness with resources as in normal operation.

Othre ideas? My config.ini is attached below.

Not that I can think of but a general read through of the parameters in the manual is always a good idea and see if something sounds like it could be valid in your case.

Rgrds Mikael

Thx.

Here's how I calculate my database's data and index memory size requirements:

PER COLUMN

Data Requirement Index Requirement Fields

-------------------------------------------------------------- 4 enum,int,time,timestamp 8 datetime size* char size + 2* varchar 256 text*** size + 25** hash key 10 size + 25** hash key + ordered key

* rounded up to nearest 4-byte boundary ** + 8 if size > 32 *** see below

PER TABLE

data subtotal = sum of Data Requirement column + 16 data records per page = (32768 - 128) / data subtotal data pages = number of records / data records per page data memory requirement = (data pages * 32768) + text requirement***

index subtotal = sum of Index Requirement column index records per page = (32768 - 128) / index subtotal index pages = number of records / index records per page index memory requirement = index pages * 32768

*** text requirement = sum of any text field data in excess of 256 bytes NOTE: This calculation may be off because it doesn't include any overhead for records, table or indexes associated with the overflow text

Is this correct? If so, then my database sizes are:

data size = 1725M index size = 360M

And using this formula:

(size * 1.1) * number of replicas / number of nodes

the minimum settings for my 2 nodes should be:

DataMemory = 1898M IndexMemory = 396M

(I've got 6GB of RAM on each of 2 NDB nodes. I'm using the hugemem Linux kernel so each NDB process can access up to 4GB--instead of 2GB as in the standard kernel.)

In the real world, though, after several days of trying, I haven't been able to run my databases with anything smaller than 2600M and 500M without getting "table full" errors. I want to understand why. Any help would be appreciated. Thx.

-- Jim Hoadley Sr Software Eng Dealer Fusion, Inc

latest config.ini:

################################################################# [ndbd default] NoOfReplicas= 2 MaxNoOfConcurrentOperations=131072 DataMemory= 2600M IndexMemory= 500M Diskless= 0 DataDir= /var/mysql-cluster TimeBetweenWatchDogCheck=10000 HeartbeatIntervalDbDb=10000 HeartbeatIntervalDbApi=10000 NoOfFragmentLogFiles=64 #TimeBetweenLocalCheckpoints=19 NoOfDiskPagesToDiskAfterRestartTUP=54 #default=40 NoOfDiskPagesToDiskAfterRestartACC=8 #default=20 #http://lists.mysql.com/cluster/1441 MaxNoOfAttributes = 2000 MaxNoOfOrderedIndexes = 5000 MaxNoOfUniqueHashIndexes = 5000

[ndbd] HostName= 10.0.1.199

[ndbd] HostName= 10.0.1.200

[ndb_mgmd] HostName= 10.0.1.198 PortNumber= 2200

[mysqld]

[tcp default] PortNumber= 2202 #################################################################

Jumpstart your cluster: http://www.mysql.com/consulting/packaged/cluster.html