February 8, 2013 at 9:53 am
Hi all
I have a table which contains login and logout times for a large set of users, and we are wanting to know how to code it to pull back the maximum number of users who are logged on at any one time during that day.
I have got no where with this at present I have searched the internet and found something here:
http://stackoverflow.com/questions/1117004/find-number-of-concurrent-users-in-a-sql-records
However I can not figure out the solution mentioned and have been unable to get it to work. If anyone has any thoughts on how best to do this I would be grateful
Thanks.
February 8, 2013 at 9:59 am
Kwisatz78 (2/8/2013)
Hi allI have a table which contains login and logout times for a large set of users, and we are wanting to know how to code it to pull back the maximum number of users who are logged on at any one time during that day.
I have got no where with this at present I have searched the internet and found something here:
http://stackoverflow.com/questions/1117004/find-number-of-concurrent-users-in-a-sql-records
However I can not figure out the solution mentioned and have been unable to get it to work. If anyone has any thoughts on how best to do this I would be grateful
Thanks.
One way would be to create a buckets table with one row representing each and every single minute of the day - is that granularity enough for you? - then, for each row on your login/logout table add 1 to all the buckets representing minutes the particular user was logged into the system.
At the end of the process just select the bucket with max() and the minute of the day represented by the winning bucket plus the value of the bucket would tell when the max() number of users was logged in and how many of them where logged in at the time.
Hope this helps.
_____________________________________
Pablo (Paul) Berzukov
Author of Understanding Database Administration available at Amazon and other bookstores.
Disclaimer: Advice is provided to the best of my knowledge but no implicit or explicit warranties are provided. Since the advisor explicitly encourages testing any and all suggestions on a test non-production environment advisor should not held liable or responsible for any actions taken based on the given advice.February 8, 2013 at 12:14 pm
The solution provided in your link is likely to be the fastest. (It essentially the same as what Paul is suggesting, but limiting the buckets to only the specific login times.) If you post what you have already tried and where you ran into problems, we can help you understand how it works.
J. Drew Allen
Business Intelligence Analyst
Philadelphia, PA
February 9, 2013 at 3:33 pm
drew.allen (2/8/2013)
The solution provided in your link is likely to be the fastest. (It essentially the same as what Paul is suggesting, but limiting the buckets to only the specific login times.) If you post what you have already tried and where you ran into problems, we can help you understand how it works.
If you're talking about Alex K's solution, it's absolutely horrible. If you take a look at the Actual Execution Plan, it has a full blown accidental CROSS JOIN in it for smaller numbers of rows and a full blown Triangular Join in it for larger numbers. I wouldn't use that code if it was the only way to get this problem done.
--Jeff Moden
Change is inevitable... Change for the better is not.
February 9, 2013 at 6:13 pm
Kwisatz78 (2/8/2013)
I have a table which contains login and logout times for a large set of users, and we are wanting to know how to code it to pull back the maximum number of users who are logged on at any one time during that day.
This was the subject of a series of articles by Itzik Ben-Gan. The fastest solution found was submitted by, among others, our very own R Barry Young. You can read all about it here:
http://www.sqlmag.com/article/tsql3/calculating-concurrent-sessions-part-3-103407
Be sure to read the whole thing, not just the first page. I have a SQLCLR solution that beats that by around 30% but unless you really need that extra bit of speed (and are quite expert with T-SQL and SQLCLR) I would stick with Barry's code.
Paul White
SQLPerformance.com
SQLkiwi blog
@SQL_Kiwi
February 9, 2013 at 10:11 pm
SQL Kiwi (2/9/2013)
Kwisatz78 (2/8/2013)
I have a table which contains login and logout times for a large set of users, and we are wanting to know how to code it to pull back the maximum number of users who are logged on at any one time during that day.This was the subject of a series of articles by Itzik Ben-Gan. The fastest solution found was submitted by, among others, our very own R Barry Young. You can read all about it here:
http://www.sqlmag.com/article/tsql3/calculating-concurrent-sessions-part-3-103407
Be sure to read the whole thing, not just the first page. I have a SQLCLR solution that beats that by around 30% but unless you really need that extra bit of speed (and are quite expert with T-SQL and SQLCLR) I would stick with Barry's code.
Freakin' awesome link, Paul. I was able to modify Barry's code to also correctly populate the MX column for the Logoffs so that I could graph the "valleys" as well as the "peaks". I've been trying to do this solution in a similar fashion and got seriously hooked because I just didn't see the 2:1 ratio that Barry included in his final formula. Thanks for posting the link. It's definitely a keeper.
Barry, if you read this post, I know it's been 3 years since you wrote the code and that article came out but thanks a million to you for writing the code and to Itzik for 'splainin' it.
--Jeff Moden
Change is inevitable... Change for the better is not.
February 9, 2013 at 10:35 pm
Jeff Moden (2/9/2013)
Freakin' awesome link, Paul. I was able to modify Barry's code to also correctly populate the MX column for the Logoffs so that I could graph the "valleys" as well as the "peaks". I've been trying to do this solution in a similar fashion and got seriously hooked because I just didn't see the 2:1 ratio that Barry included in his final formula. Thanks for posting the link. It's definitely a keeper.
Yes, it's very clever but quite simple at the same time, once the concepts sink in. Once SQL Server supports proper ordered aggregates, the problem will be trivial.
Paul White
SQLPerformance.com
SQLkiwi blog
@SQL_Kiwi
February 10, 2013 at 12:07 am
I've worked with the "difference between ROW_NUMs" a lot and, like you say, once you've got the concept down, it's very simple. Heh... unless you did like I originaly did and miss the bloody 2:1 ratio that Barry used in his final calculation.
Thanks again, Paul.
--Jeff Moden
Change is inevitable... Change for the better is not.
February 10, 2013 at 1:56 pm
Are you all set now or do you need some additional help?
--Jeff Moden
Change is inevitable... Change for the better is not.
February 10, 2013 at 2:02 pm
Crickey guys thanks very much for all the replies, I will delve properly into them tomorrow when back at work, I decided to give myself a weekend off this week and took some R&R, but will definitely post back if I get stuck further.
Thanks again
February 10, 2013 at 6:44 pm
Jeff Moden (2/9/2013)
SQL Kiwi (2/9/2013)
Kwisatz78 (2/8/2013)
I have a table which contains login and logout times for a large set of users, and we are wanting to know how to code it to pull back the maximum number of users who are logged on at any one time during that day.This was the subject of a series of articles by Itzik Ben-Gan. The fastest solution found was submitted by, among others, our very own R Barry Young. You can read all about it here:
http://www.sqlmag.com/article/tsql3/calculating-concurrent-sessions-part-3-103407
Be sure to read the whole thing, not just the first page. I have a SQLCLR solution that beats that by around 30% but unless you really need that extra bit of speed (and are quite expert with T-SQL and SQLCLR) I would stick with Barry's code.
Freakin' awesome link, Paul. I was able to modify Barry's code to also correctly populate the MX column for the Logoffs so that I could graph the "valleys" as well as the "peaks". I've been trying to do this solution in a similar fashion and got seriously hooked because I just didn't see the 2:1 ratio that Barry included in his final formula. Thanks for posting the link. It's definitely a keeper.
Barry, if you read this post, I know it's been 3 years since you wrote the code and that article came out but thanks a million to you for writing the code and to Itzik for 'splainin' it.
Truly an amazing solution for this problem.
+10 to Barry and Itzik!
My thought question: Have you ever been told that your query runs too fast?
My advice:
INDEXing a poor-performing query is like putting sugar on cat food. Yeah, it probably tastes better but are you sure you want to eat it?
The path of least resistance can be a slippery slope. Take care that fixing your fixes of fixes doesn't snowball and end up costing you more than fixing the root cause would have in the first place.
Need to UNPIVOT? Why not CROSS APPLY VALUES instead?[/url]
Since random numbers are too important to be left to chance, let's generate some![/url]
Learn to understand recursive CTEs by example.[/url]
[url url=http://www.sqlservercentral.com/articles/St
February 15, 2013 at 3:50 am
I just wanted to let you all know that I implemented the solution and it works a treat as the previous poster says many thanks to Barry and Itzik.
February 15, 2013 at 11:20 am
Perhaps a bit late but a different solution in one statement 🙂
To set the data up based on the solution of http://www.sqlmag.com/article/tsql3/calculating-concurrent-sessions-part-3-103407:
USE tempdb;
IF OBJECT_ID('dbo.Sessions', 'U') IS NOT NULL DROP TABLE dbo.Sessions;
CREATE TABLE dbo.Sessions
(
keycol INT NOT NULL,
app VARCHAR(10) NOT NULL,
usr VARCHAR(10) NOT NULL,
host VARCHAR(10) NOT NULL,
starttime DATETIME NOT NULL,
endtime DATETIME NOT NULL,
CONSTRAINT PK_Sessions PRIMARY KEY(keycol),
CHECK(endtime > starttime)
);
GO
CREATE INDEX idx_nc_app_st ON dbo.Sessions(app, starttime) ;
CREATE INDEX idx_nc_app_et ON dbo.Sessions(app, endtime);
GO
--- Populate the table
declare @i int = 1
declare @DT_Rnd datetime
while @i < 1000
begin
Set @i = @i + 1
set @DT_Rnd = dateadd( mi , RAND()* 1440 , cast('20090212' as datetime) )
INSERT tempdb.dbo.Sessions(keycol, app, usr, host, starttime, endtime)
VALUES( @i
, 'app' + right( '00' + CAST ( 1 + cast( RAND()* 15 as int) as varchar(2)) , 2)
, 'user' + right('000' + CAST ( 1 + cast( RAND()* 150 as int) as varchar(2)) , 3)
, 'host' + right('000' + CAST ( 1 + cast( RAND()* 240 as int) as varchar(2)) , 3)
, @DT_Rnd
, dateadd( mi , 5 + (RAND()* 50) , @DT_Rnd )
);
And for the single statement solution:
Select APP
, [No of Concurrent users]
, [Point in Time]
from (
Select Toe.app
, Toe.[Point in Time]
, [No of Concurrent users] = COUNT(distinct keycol)
, RID = row_number() over ( partition by toe.App order by COUNT(distinct keycol) desc )
from ( -- Time of Events
select app , [Point in Time] = starttime from tempdb.dbo.Sessions
union Select app , [Point in Time] = endtime from tempdb.dbo.Sessions
) as TOE
inner join tempdb.dbo.Sessions as S1
on s1.App = Toe.app
and Toe.[Point in Time] >= s1.starttime
and Toe.[Point in Time] < s1.Endtime
group by Toe.app , Toe.[Point in Time]
) as c
where RID = 1
order by App , RID
Polite comments are welcome
Viewing 13 posts - 1 through 12 (of 12 total)
You must be logged in to reply to this topic. Login to reply