GS 2017 Setup: Performance

Hi,

i have simple point for discussion. How did you set your infrastructure? Do you have separate machines for each component? For example, Application Server, RabbitMQ server, DB server or do you host RabbitMQ on same machine as Application server? Also, do you see any difference in performance, does project preparation time differs for instance? How is the performance in comparison with GS2015?

Our server specifications: Vast majority of our servers are VMs and usual specs are App. server 8 Cores and 16GB RAM; SQL server 24 Cores and 65GB RAM. Standard Virtual SCSi disk drives. 

RabbitMQ on separate machine: We tried both variants and did not notice any significant difference, we are not entirely sure why though. Maybe the another network communication could be the bottle neck so the advantage of having separated RabbitMQ and Application server is not visible.

I thought it would be good to discuss those points here with wider audience as i have not found anywhere in official SDL sources any details about it, neither about server specs, the details provided are bit vague in my view.

Thank you for sharing your thoughts.

Regards,

Fana

  • Hi,

    we have separate machines for sql server and "everything else" (including rabbit). We didn't achieve any specific performance improvement by trying different settings.

    The only thing that seems to get faster project creation (referred specifically to pre-translate phase) is periodically reindex of translation memories, we programmed that via the rest api connected to our main management system.

    GS2015 used to perform better IMHO, of course the complexity of GS2017 is significantly higher, this probably explains why.

    Also using more than one TM in project creation really slows down things, as slows down having many small files instead of few bigger ones, this has more to do with the process revolving around Studio, than Studio itself.

    I hope this may be of any help, I agree with you: details are a bit vague and sharing these points with a bigger audience would help, since trying and testing different configurations when it comes to servers is not that easy.

    Regards to you!

  • Hi Frantisek,

    When it comes to Project preparation and TM speed, the single biggest factor is SQL Server performance. I.e. we had a system where the creation of a reference project took around 20 minutes, after upgrading the SQL Server hardware including disk IOPS. The project creation time was brought down to 8 minutes, without any software code changes.
    We also strongly recommend the SQL Server to be on a separate machine to ensure not congestion with the application hardware.

    Separating RabbitMQ would not help in this instance, as that's typically not involved in project creation or TM lookups.

    In addition to this, CU08 will bring improvements when it comes to performance.

    Thanks,
    Luis