The relationship between tempdb and replication slow response alert.

It's your complex Merge Replication with CDC?
Do you have the exact error message?

no.

it just merge replication slow and I am trying to see if enhancing tempdb can help.

I guess you should follow the usual recommendations for tempdb:

• Isolate tempdb in a dedicated and fast disk
• Have as many data files as the server has processors, till a maximum of 8
• Having all files with the same size

hi,

yes, I have all that.

but one thing:

"Have as many data files as the server has processors, till a maximum of 8"

 that one should be only for SQL 2016 and above.

we all talking about .mdf and .ndf but how about log ? all data writes to log first and then committed to the data file.

how about as much tempdb log file as tempdb data file? and all equal size ?

this might also solve the tempdb log contention problem.

that one should be only for SQL 2016 and above.

No. That's a rule that DBAs are using since at least SQL Server 2005.

we all talking about .mdf and .ndf but how about log ? all data writes to log first and then committed to the data file.

 how about as much tempdb log file as tempdb data file? and all equal size ?

The rule is only for data files. There's no sense to create more that a transaction log file since it's sequential writing (the engine can't write in more than a transaction log file at a time).

"No. That's a rule that DBAs are using since at least SQL Server 2005."

what I know there are only 3 x rules for tempdb:
1) number of tempdb file = number of CPU core.
2) number of tempdb file = 1/2 number of CPU core.
3) number of tempdb file = number of CPU socket,

I know in SQL 2016, default number of tempdb file is 8 and we no need to calculate any more, and just keep adding tempdb if tempdb contention is  found.

that's why I don't' understnad what you said.

"There's no sense to create more that a transaction log file since it's sequential writing (the engine can't write in more than a transaction log file at a time)."

tempdb data file is parallel write ?

what I know there are only 3 x rules for tempdb:
 1) number of tempdb file = number of CPU core.
 2) number of tempdb file = 1/2 number of CPU core.
 3) number of tempdb file = number of CPU socket,

So if you have for example a server with 4 cores in 2 sockets, how many files would you create?

I know in SQL 2016, default number of tempdb file is 8 and we no need to calculate any more, and just keep adding tempdb if tempdb contention is  found.

This is for facilitate the life on non-DBAs :) And I bet that you don't see there more than one transaction log file, do you?

tempdb data file is parallel write ?

Any SQL Server database is, since it has more than one data file.

"So if you have for example a server with 4 cores in 2 sockets, how many files would you create?"

it is a total of 8 cores, so 8 tempdb files, but I am more worry about the tempdb log files, as data write to that first.

"This is for facilitate the life on non-DBAs :) And I bet that you don't see there more than one transaction log file, do you?"

yes, it is for not that experts DBA.  

sorry you lost, I see a lot of situation where we have more than one log file and that's why I ask if it really helps.

you know now if I run who is active, you can see query from time to time write to tempdb log, this make me worry/think about one more tempdb log files to speed up the tempdb operation.

"Any SQL Server database is, since it has more than one data file."

just the log is not ...

do you know if it is in a round robin manner ?

"This is for facilitate the life on non-DBAs :) And I bet that you don't see there more than one transaction log file, do you?"

can you tell me why people would like more than 1 tempdb log files ?

can you tell me why people would like more than 1 tempdb log files ?

Besides you and don't know anybody that like to have more than 1 tempdb log file. The only situation that I can think that justifies a second transaction log file is when the disk is full and for emergency you'll need to create a 2nd transaction log file in another disk for an immediate workaround.

Vitor Montalvão,

"You're only losing time if trying to do something that's not possible."

ok, this is good answer then..

I just see TEmpdb contention message in replicaiton and I am not sure if I use create more log file instead of data files.

"The only situation that I can think that justifies a second transaction log file is when the disk is full and for emergency you'll need to create a 2nd transaction log file in another disk for an immediate workaround."

for emergency situation, like what we have today, we just kill spid and shrink log.

then tempdb fine.

tks victor.