Typical Latency In Transactional Replication

Jul 31, 2007

I have been working on a project to evaluate transactional replication in SQL Server 2005 to find out if it will be suitable to provide real time reporting without impacting the primary database server. Thus far I have not seen latency better than several seconds replicating a change to a table or by utilizing the "tracer tokens" troubleshooting functionality. It usually takes a couple seconds from the publisher to the distributor and a couple more from the distributor to the subscriber.

I guess my question is actually whether it is possible to achieve sub-second (milliseconds) latency for transactional replication under any circumstances. Someone on another forum told me it was not, so I thought I would ask a separate community to confirm. Regarding the hardware, the latency on ping is less than 1ms between the servers and the servers themselves are decently powered quad processor boxes w/ 4GB of ram each. It seems to have the same latency of 3-5 seconds regardless of whether I publish a single table or all of the tables in the database. I currently have it configured with the publisher on one server and the distributor/subscriber on the second with the log reader and distributor agents set to run continuously. If it is possible, please recommend configuration of the agents or the topology of the PDS to achieve this. Thanks in advance for any insight you provide.

Dan

View 2 Replies

Monitor Transactional Replication Latency

Jun 12, 2008

CREATE PROCEDURE [dbo].[proc_replication_latency_monitor]
AS

-- Procedure created by Henning Frettem (Lumbago), www.sqlteam.com, 12th of June 2008

BEGIN

SET NOCOUNT ON

DECLARE
@PublishedDBName varchar(200),
@SQL nvarchar(2000),
@PublicationPK int,
@PublicationName varchar(200),
@PublicationDatabase varchar(200),
@TracerTokenID int,
@Parmameters nvarchar(500)

DECLARE @Publications table (
PublicationPK int IDENTITY(1, 1) PRIMARY KEY CLUSTERED,
PublicationDatabase varchar(200),
PublicationName varchar(200),
TracerTokenID int,
TracerTokenPostTime datetime
)

DECLARE @Latency table (
PublicationPK int,
DistributorLatency bigint,
Subscriber sysname,
SubscriberDB sysname,
SubscriberLatency bigint,
OverallLatency bigint
)

--> Cursor for fetching all publications in all databases
DECLARE curPublishedDatabases CURSOR LOCAL FAST_FORWARD FOR
SELECT [name] FROM sys.databases WHERE is_published > 0

OPEN curPublishedDatabases
FETCH curPublishedDatabases INTO @PublishedDBName
WHILE @@FETCH_STATUS = 0
BEGIN
SET @SQL = N'SELECT ''' + @PublishedDBName + ''', [name] FROM ' + @PublishedDBName + '.dbo.syspublications'
INSERT INTO @Publications (PublicationDatabase, PublicationName)
EXEC sp_executesql @SQL

FETCH NEXT FROM curPublishedDatabases INTO @PublishedDBName
END

CLOSE curPublishedDatabases
DEALLOCATE curPublishedDatabases

--> Cursor for posting tracer tokens
DECLARE curPublications CURSOR LOCAL FAST_FORWARD FOR
SELECT PublicationPK, PublicationDatabase, PublicationName FROM @Publications

OPEN curPublications
FETCH curPublications INTO @PublicationPK, @PublicationDatabase, @PublicationName
WHILE @@FETCH_STATUS = 0
BEGIN
SET @Parmameters = N'@PublicationName varchar(200), @TracerTokenID_OUT int OUTPUT';
SET @SQL = N'EXEC ' + @PublicationDatabase + '.sys.sp_posttracertoken @publication = @PublicationName, @tracer_token_id = @TracerTokenID_OUT OUTPUT'
EXEC sp_executesql @SQL, @Parmameters, @PublicationName = @PublicationName, @TracerTokenID_OUT = @TracerTokenID OUTPUT

UPDATE @Publications SET
TracerTokenID = @TracerTokenID,
TracerTokenPostTime = GETDATE()
WHERE PublicationPK = @PublicationPK

FETCH NEXT FROM curPublications INTO @PublicationPK, @PublicationDatabase, @PublicationName
END
CLOSE curPublications
DEALLOCATE curPublications

--> Wait two minutes for all tokens to be commited at all subscribers
WAITFOR DELAY '000:02:00.000'

--> Then check the results for each posted token
DECLARE curTokens CURSOR LOCAL FAST_FORWARD FOR
SELECT PublicationPK, PublicationDatabase, PublicationName, TracerTokenID FROM @Publications

OPEN curTokens
FETCH curTokens INTO @PublicationPK, @PublicationDatabase, @PublicationName, @TracerTokenID
WHILE @@FETCH_STATUS = 0
BEGIN
--> Insert token history for each token
SET @SQL = N'EXEC ' + @PublicationDatabase + '.dbo.sp_helptracertokenhistory @publication = ''' + @PublicationName + ''', @tracer_id = ' + CAST(@TracerTokenID AS VARCHAR(50)) + ''
INSERT INTO @Latency
(DistributorLatency, Subscriber, SubscriberDB, SubscriberLatency, OverallLatency)
EXEC sp_executesql @SQL

--> Make sure that the PublicationPK is added to the token history
UPDATE @Latency SET PublicationPK = @PublicationPK WHERE PublicationPK IS NULL

--> Clean up the tracer token
SET @SQL = N'EXEC ' + @PublicationDatabase + '.sys.sp_deletetracertokenhistory @publication = ''' + @PublicationName + ''', @tracer_id = ' + CAST(@TracerTokenID AS VARCHAR(50)) + ''
EXEC sp_executesql @SQL

FETCH NEXT FROM curTokens INTO @PublicationPK, @PublicationDatabase, @PublicationName, @TracerTokenID
END
CLOSE curTokens
DEALLOCATE curTokens

SELECT
b.PublicationDatabase,
b.PublicationName,
a.Subscriber,
a.SubscriberDB,
a.OverallLatency,
b.TracerTokenPostTime
FROM @Latency a
INNER JOIN @Publications b
ON a.PublicationPK = b.PublicationPK

END

--
Lumbago

View 7 Replies View Related

Troubleshooting Transactional Replication Latency

Oct 19, 2007

Hello!

We have setup up transactional replication with dedicated distributor in SQL Server 2005 environment. I have noticed that during particular time of the day latency is increasing dramatically. I have been checking Tracer Tokens and Total latency during that time is around 30-40 min (both publisher to distributor and distributor to subscriber is taking much longer that normal). Normally, it is less than 10 sec. I was wondering if there is a way to pinpoint exact cause of the latency. This is pull subscription.
I would appreciate if someone can share (or point to the right direction) best practice on transactional replication setup/maintenance. My understanding is that only committed transactions are replicated, correct? I checked database on publisher and didn't see any outstanding long running transaction.

Any help is greatly appreciated.

Thanks,
Igor

View 3 Replies View Related

Commiting Transactions/Truncating Transactional Logs/Replication Latency

Feb 22, 2008

Hi,

I have multiple questions and would appreciate any suggestions in resolving them. I'm a novice to these issues.

1) First of all, what is the exact command to setup the trunc. log on chkpt. option on for a transactional log of a SQL Server 2000 database? Is this option on by default? I have noticed for one of the databases I'm managing that the transaction log was over 12 GB, while the db was only 425 MB.

2) How's it possible to run a DBCC TRACEON to see the content of the transaction log to see if we are having any issues with any uncommitted transactions, i.e. updates, inserts, and/or deletes.

3) What are the commands to truncate inactive transactions and increasing the readbatchsize?

4) lastly, how do I validate transactional replication via checksum and find valid latency between a small number of changes that need to be committed between publisher and subscriber.

Thank you so much in advance!
Sincerely,
Alla

View 3 Replies View Related

Transactional Replication Over Satellite Connection. Set Parameters For High Latency?

Jul 31, 2007

I have 2 servers; 1 in the head office and the 2nd at a remote location connected over a satellite. The satellite connection has typical latency and bandwidth. 2 mb down and 128kb up. Latency is average of 700ms.

Both servers have plenty of processing power and RAM and are very underutilized.

I have 1 push publication at the publisher at the remote location. The publication has 1 article. The table that is contained in the article only grows; ie inserts are the only statements applied, no updates or deletes. The table grows by 5 inserts every second(approximate). the table has 5 columns. none of the columns are blobs etc. An ID field, 1 datetime, 2 integers and 1 float column.

I have the agents set for continuous with a 2 second polling interval.

I am aware of the following command parameters for the agents but not sure which way to go to maximize the throughput of the data.

MaxCmdsInTran
SubscriptionStreams
ReadBatchSize
CommitBatchSize
PollingInterval

The subscriber table also has an insert after trigger to update another small table with the current value. Under normal network conditions, the trigger works great. Over the sat connection with high latency, the smaller target table seems to get a lock placed on it. I can only guess that the lock is from the inserts and perhaps a large number of inserts per transaction?

Under normal network conditions, the latency between publisher - distribution - subsciber is about 4 seconds.

What would be good command parameter settings for the agents to compensate for the latency?

Thanks,
Kevin

View 5 Replies View Related

Replication Latency

Feb 12, 2007

hi

I have a transactional replication that was running fine for about few months, but eversince there was a massive update of records at the publisher site and when I use the replication monitor, i notice Status: Performace Critical, Subscription:[localservername]:[Orders_Repl], Performance:Critical, Latency:03:34:47.

I was thinking of redoing the whole replication but before i am about to proceed. Can you guys explained what happen and what does it meant when Latency:03:34:47. I don't remember seeing a Latency:03:34:47.

I am hoping to solve this critical performance issue hope you guys can help. Thanks