From 760b213bbf7ac7bf41c932a13488607f363c398e Mon Sep 17 00:00:00 2001 From: Alek Kundla <129544037+akundla-google@users.noreply.github.com> Date: Fri, 23 Feb 2024 17:52:11 -0600 Subject: [PATCH] Update system memory guidelines in load testing guide Need to reflect http://go/b&a-roma-mem-calc, which says that each roma worker can require up to 1.92 GB of memory, and this doesn't even include any requests. Old wording was confusing, may have recommended either a high-cpu configuration or a high-mem configuration depending on your interpretation. --- tools/load_testing/README.md | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/tools/load_testing/README.md b/tools/load_testing/README.md index 89eafb26..5b20420f 100644 --- a/tools/load_testing/README.md +++ b/tools/load_testing/README.md @@ -26,9 +26,11 @@ Note the following while building and deploying the services: - Modify the terraform configuration based on the load(QPS) you are testing with: - **bidding_machine_type, bfe_machine_type, auction_machine_type, sfe_machine_type**: Specify - the size of the machine for all the services. Note: SFE can use a standard machine while - BFE, Bidding and Auction Services will need a high memory configuration. Bidding and Auction - are usually recommended to be more compute optimized. + the size of the machine for all the services. Note: SFE can use a standard machine. + BFE is typically more memory-insentive than SFE. The Bidding and Auction Services will + need at least 3 GB of memory per Roma worker. + Bidding and Auction workloads are typically relatively more compute-intensive so are + usually recommended to have relatively more compute power. - **Min_replicas_per_service_region** - **Max_replicas_per_service_region** - **JS_NUM_WORKERS** : Specify the number Roma workers (equal to the vCPUs in the machine)