How many concurrent users can a GSA handle?

Summary: How many concurrent users performing searches can a single search appliance handle?

Cause: Many administrators try to plan their search solution around the number of users. However, from the search appliance point of view, the number of users is not a very conclusive number. The search appliance is built to handle incoming requests as fast possible; ideally, in sub second intervals. Therefore, multiple users can use the same second to perform a search.

In summary, it does not count how many users are using GSA. It is more important how many queries per second (QPS) or minutes as a user can submit multiple queries at any time. For example, 50 concurrent queries is the maximum for G100 and G500.

If each query only takes 0.5 seconds to finish, in this case, GSA can handle 100 queries per second (QPS). Of course, normally, that is not the case. You should find out the average QPS for a normal day and use that number to estimate how many GSAs you need for handling serving.

How to list your recent QPS:

You can find out QPS that by running a 'Search Log' report over a period of time (no more than 90 days as that is the limit how long your GSA will save the search logs). Once you have the search log, here are two UNIX commands that will provide you QPS and response time:

cat <search_log_filename> | awk '{ print $4, $6, $7, $8, $9, $10, $11, $12 }' | cut -b 2-21 | sort -n | uniq -c | sort -nr | head

This will provide you the Top 10 QPS (you can change the return number by change this 'head -20' to get top 20 results).

cat <search_log_filename> | awk '{ print $12, $4 "]"}' | sort -n

This will sort the search log according to response time in the log.

If you need slightly more QPS:

Implement a reverse proxy that allows you to queue requests sent to the appliance or distribute the queries between multiple appliances. You can also try to use less costly queries (i.e. without query expansion, different sorting, etc.).

Additional Information:

Was this helpful?
How can we improve it?