June 2, 2009 at 5:17 am
We have a customer set up with sql running from a cluster. The main publisher is also the distributor. The publisher has 55 subscribers on WAN over broadband. Replication is merge type and push from publisher with conflict resolution as first to publisher wins.
A typical site can complete replication in around 40 seconds. All servers are the same specification and have the same low contention broadband package.
Some of the subscribers can take hours to perform the same replication as other sites (worst case was 14+ hours). It is always the same sites.
As an example:- one of these is currently taking 28:53 mins, with 68 uploaded changes and 355 downloaded.
upload to publisher 99% of total, duration 00:28:40
inserts 57, update 11, delete 0, conflicts 0
Heavy upload articles:
Article1 22% load, 00:06:21 , 1 insert
Article2 53% load, 00:15:23, 1 insert
Download to subscriber 1% of total, Duration 00:00:32
inserts 248, updates 50, deletes 21, conflicts 1
Heavy Articles
Article3 0% load, 00:00:03, 40 inserts, 20 deletes
Article4 0% load, 00:00:02, 1 conflict
Summery -
Out of all the time taken, 99% of the time is used pushing info from the branch to the main server.
53% of the total time taken is used inserting 1 record in to 1 table. 15 mins and 23 seconds.
The broadband has been toughly checked and with upload and download tests of large files and comparing this to healthy sites.
Another strange feature is that the upload bandwidth for these unhealthy sites is extreamly high often using all of the upload bandwidth when communicating to the publisher, What data is in this bandwidth? why does it cause the inserts to take so long?
June 3, 2009 at 9:15 am
What I really don't understand is how can some subscribers add data to the publishers articles with no problems, and yet for this number of site they always seam to have consistant problems inserting to the publisher???
Does anyone else have any experience of this, or know where we could look for causes?
June 12, 2009 at 3:24 am
Could it be possible that the subscriber is having trouble inserting up to the publisher on selected articles at selected sites?
for example if i have an article with 1 row to upload to the publisher, and this 1 row takes 15 mins, along with massive upload bandwidth could it be resending the 1 row 1000/ 10000 times or more before its actually uploaded? What else could account for the extra bandwidth.
Does any one have any ideas on how I can move forward with this.
Or recommend a replication expert consultant we can pay to get this resolved.
Viewing 3 posts - 1 through 3 (of 3 total)
You must be logged in to reply to this topic. Login to reply