Your own workload

The Quickstart starts from a prebuilt workload. To simulate another workload, prepare the disk and the snapshot yourself in three steps, then continue with the Quickstart's Init Warm:

1. Create an empty disk image;
2. Boot it, install the operating system and the workload, and save the snapshot boot;
3. Load boot, start the workload, and save the snapshot loaded, in which it runs in steady state.

The args file

Without a prebuilt workload, qflex.args describes the machine and the workload itself:

--core-count 8
--double-cores
--quantum-size-ns 2000
--llc-size-per-tile-mb 2
--parallel
--network none
--memory-gb 32
--workload-name web-search
--population-seconds 5
--experiment-name web-search-8c-001
--image-name root.qcow2
--working-folder <WORK_DIR>
--no-unique
--use-image-directly

• --core-count is the number of simulated cores: 1, 2, 4, 8, 16, 32, 48 or 64.
• --double-cores gives the VM twice as many vCPUs: the second half runs the workload's client, as the phantom core group.
• --use-image-directly makes the experiment use <WORK_DIR>/images/<image name> itself instead of a copy, so the snapshots saved in the steps below stay in your image.
• Keep the machine settings (--core-count, --double-cores, --memory-gb, --network) the same in every step: a snapshot only restores on the machine it was taken on.

As in the Quickstart, write absolute paths: xargs does not expand ~ or $VARIABLES.

Core groups

Without --groups, the measured cores form one group named after --workload-name (IPC 2), plus the phantom group (IPC 4) with --double-cores. To describe a consolidated workload or other IPC estimates, give the groups as name:cores[:ipc]:

--groups "dc:0-7:0.6,ws:8-15:4.2,phantom:16-31:4"

• The measured groups must cover the cores 0…N-1 exactly once, for --core-count N.
• With --double-cores, the phantom group must be exactly the second half of the vCPUs, N…2N-1.
• The IPC is optional: 2 for a measured group, 4 for phantom.

See Core groups in the Quickstart for what the groups are used for. --groups also overrides the groups of a prebuilt workload.

1) Create the base image

This creates an empty qcow2 disk <WORK_DIR>/images/root.qcow2 (8 GB; change it with --image-size-gb). Use the same --working-folder and --image-name as in qflex.args.

2) Boot and install the workload

--use-cd-rom attaches the Alpine Linux installer (downloaded on first use). Log in as root without a password, install Alpine onto the disk (setup-alpine), then install your workload and its dependencies. Later boots of the disk do not need --use-cd-rom.

When the VM is in the state you want to keep, save it from the QEMU monitor:

1. Press Ctrl+A, then C. The prompt becomes (qemu).
2. Save the snapshot:

   (qemu) savevm boot
   

3. Quit with (qemu) quit.

3) Load and start the workload

This restores boot with one clock per core, as in the later steps. Start the workload's components inside the VM, wait until it runs in steady state, and save the snapshot from the QEMU monitor:

(qemu) savevm loaded

Continue with the Quickstart

Run Init Warm from your snapshot, then follow the Quickstart from there:

Make sure the workload is running in loaded: Init Warm waits for the workload to touch the caches, so an idle VM makes it run for a long time.