You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Renamed GPU-related flags like so (old_flag_value -> new_flag_value):
gce_gpu_count -> gpu_count
gce_gpu_type -> gpu_type
Replaced object_storage_objects_written_file* flags with prefix versions.
Renamed tf_version flag to tf_pip_package flag to allow the user to specify
the full tensorflow pip package name to be installed, instead of just being
able to control the version.
Changed default Tensorflow package when using CPUs to an Intel CPU-optimized
build.
Renamed cuda_toolkit_8 to cuda_toolkit
Migrated cluster boot benchmark default machines to 'default_dual_core'.
Changed metric name in mnist and inception3.
Renamed the tf_batch_size flag in tensorflow_benchmark to tf_batch_sizes.
Updated GCP sole tenancy support. Removed gcp_host_type added gcp_node_type.
Fixed missing installation directory in gpu_pcie_bandwidth benchmark.
Removed duplicated metrics from the HPCC benchmark. Metrics now use the name
from the summary section of the benchmark-produced output.
Benchmarks are expected to not modify FLAGS in any way. If FLAGS are modified
and multiple run configurations are run in a single PKB invocation, benchmark
configurations may be incorrect.
Changed TF Serving benchmark to use ResNet instead of Inception.
Renamed prepare_sleep_time flag to after_prepare_sleep_time.
multichase_taskset_options flag changed to multichase_numactl_options.
Deprecated CUDA version 8.
Remove support for supplying CUDNN package path via the 'cudnn' flag.
Raise IssueCommandError by default when vm_util.IssueCommand return code is
non-zero. Previous behavior can be emulated by setting raise_on_failure=False.
Fix AliCloud support by using the newer aliyun (instead of aliyuncli) cmd
line utility and removing Paramiko dependency
Changed support for reusing buckets for Azure blob service benchmarking. The
storage account and resource group are now named based on the bucket name so
that subsequent runs can use the same bucket.
Changed Windows Server versions for GCP and Azure. Before this AWS used
Windows Server Core (non-graphical), while GCP and Azure used Windows Server
Base (graphical). All existing windows os_types now refer to the Core
version and there is a new 'windows*_base' os_types for users who want to RDP
in and use a full GUI.
Changed --aws_user_name default from ubuntu to ec2-user. Individual
OS types including Ubuntu should still override it appropriately.
New features:
Windows benchmarks can now be run from linux controllers.
MXNet benchmarks can now be run from linux controllers.
Added initial support for preprovisioning benchmark binaries in the cloud,
if binaries are not located in local /data directory.
YCSB benchmark for Cloud Redis in GCP, Elasticache Redis in AWS, and
Redis Cache in Azure.
YCSB benchmark for DynamoDB in AWS.
Added a flag, run_stage_iterations, which causes a benchmark's run stage to be
called a specified number of times
Added cuda_toolkit_version flag
Added support for CUDA Toolkit 9.0
Added MXNet support for CUDA Toolkit 9.0
Added new version of HPCG, with CUDA 9 support
Added latest gen 'default_dual_core' machine type configuration for AWS, Azure and GCP.
Added memory as disk type for Linux only.
Added support for publishing individual dstat samples with --dstat_publish_regex.
Added Inception v3 benchmark that supports CPU, GPU and TPU. This benchmark is
the same as TensorFlow with inception3 for CPU and GPU.
Added container image building.
Added netperf container benchmark.
Added AWS container registry (ECR).
Added Azure container registry (ACR).
Added Google container registry (GCR) and added GKE autoscaling.
Added create_time to VM metadata.
Added new ycsb workload where each payload is 1mb versus the default 1kb.
Added Tensorflow Serving benchmark which tests the throughput and latency of the
standard model-server using a pre-trained inception3 model.
Added AWS Fargate support.
Added the ability to pass arbitrary parameters to tf_cnn_benchmarks.py in the
Tensorflow benchmark, through the benchmark_args flag.
Added hdrhistogram support to ycsb package.
Added support for custom machine types on GKE.
Added container_cluster_version flag to container_service.py.
AWS EFS support via "disk_type: nfs"
Added --disk_fill_size and --after_prepare_sleep_time flags
Add timeout_minutes flag to assist with cleaning up stale resources
All AWS resources and Azure resource groups are now tagged. (including timeout_minutes value).
Added windows udp test using iperf3.
Added timeout-decorator python package.
Added timeout support for windows vm remote command call.
Added IOR benchmark.
Added mdtest to IOR benchmark.
Added Elastic Container Service (EKS) as a container cluster type.
Added ResNet benchmark.
Added support for ACI (Azure Container Instances).
Added spec cpu 2017 and feedback optimization for peak runs.
Added glibc benchmark.
Added lmbench benchmark.
Added support to --disable_interrupt_moderation for Windows VMs on AWS.
Added support to --disable_rss for Windows VMs on GCP.
Added act benchmark.
Added --gce_tags flag to add --tags when launching VMs on GCP.
Added PKB support to publish samples immediately.
Adding benchmarking of Memcached on cloud VMs using memtier benchmark.
Adding support for different client/server machine types on memcached-memtier.
Adding functionality in memtier benchmark to run a variable number of threads
counts in each run.
Added support for capacity reservations for VMs and added AWS implementation.
Added support for adding additional flags to mount and /etc/fstab.
Added support for Windows2012, 2016, and 2019 for AWS, Azure, and GCP.
Terasort implementation on dpb backend.
Added cluster boot benchmark implementation on dpb backend.
Support multiple redis versions in cloud redis.
Azure Files support via "disk_type: smb"
Added MLPerf benchmark.
Added stress-ng benchmark.
Added flag --after_run_sleep_time which instructs PKB to sleep for the number
of seconds specified after the run stage has completed.
Added P5 cache type for Azure managed redis.
Added AWS elasticache Memcached provider.
Added additional options to the stress-ng benchmark.
Added support for launching GCE VMs into the standard network tier.
Added a demo under tools/.
Added Nginx benchmark.
Added support for measuring VM reboot time to the cluster boot benchmark.
Added Cygwin support to WindowsVirtualMachine and enabled a Windows+Cygwin
version of Coremark.
Added sar trace utility to measure cpu steal time.
Added ssh_via_internal_ip flag to use internal IP addresses when ssh'ing to
runner VMs, will use external IP addrs if not set.
Added support for launching GCP VM with shielded VM secure boot
Added Amazon Linux 2 os type. --os_type=amazonlinux2
Redis enterprise benchmark.
Added flag --append_kernel_command_line, which allows appending to the
kernel command line when using Linux os types.
Added Spark SQL benchmark.
Added aws_dynamodb_connectMax flag to vary the maximum number of connections
to dynamodb.
Added helpmatchmd flag to dump markdown formatted help strings
Enable specifying source ip when creating a firewall rule.
Added support for T4 GPUs on GCE.
Added mpstat utility to measure cpu processor stats.
Added a benchmark to run spark application on a cluster that computes an approximation to pi.
Added a benchmark to run spark io workload on a cluster that provisions a database and queries it.
Added flag --record_lscpu=True to record lscpu data as its own metric "lscpu"
Added support for the Clear Linux OS type for AWS and static VMs (e.g. --os-type=clear)
Added flag --placement_group_style to allow or disable different AWS placement groups options for ec2 VM benchmarks.
Flag supports cluster, spread, and none placement group options.
Added support for availability zones for Azure. Format for availability zone support is "location-availability_zone".
Example: eastus2-1 specifies Azure location eastus2 with availability zone 1.
A PKB zone can be either a Azure location or an Azure location with an availability zone.
Added resource class and mapping for AWS Athena Service.
Added support for Azure Dedicated Hosting. Simply add flag --dedicated_hosts=True to use.
Records two new samples for /proc/cpuinfo data.
Added support for specifying a number of vms per host for Azure & AWS Dedicated Hosting.
Simply add flag num_vms_per_host=<#> to use. Benchmark fails if the # of vms per host specified exceeds the memory capacity of the host.
Added support for skipping the creation of Azure availability sets with the --azure_availability_set=False flag (fixes #1863)
Added support for AWS EFA networking with --aws_efa=True flag.
Added NCCL benchmark for GPU networking.
Added AWS DAX provider.
Added Google Cloud Firestore ycsb benchmarks.
Added support for un-managed data processing yarn cluster benchmarking.
Added support for UDP_STREAM tests to netperf_benchmark
Added flags to select stream size to netperf_benchmark
Added placement group flag support for Azure. "cluster" will create proximity placement group.
"spread" will create an availability set.
Added a tcpdump collector with --tcpdump flag.
Added ability to load datasets into BigQuery from csv files on GCS.
Updated AzureVirtualNetwork to support location level virtual nets, which allows support for cross zone benchmark runs.
Enhancements:
Support for ProfitBricks API v4:
Add profitbricks_image_alias flag and support for image aliases
Add new location, us/ewr
Add aws_image_name_filter flag to ease specifying images.
Add c5/m5 support for NVME disks.
Add MNIST benchmark support for TPU, CPU and GPU
Created KubernetesPodSpec which allows the user to specify kubernetes resource
requests and limits, including fractional CPUs.
Add skip_pending_runs_file flag and extension hooks to workaround SIGINT issues.
Add support for specsfs2014_load parameter as an integer list.
Publishers can be extended through external modules.
Add run_processes_delay flag to stagger parallel runs.
Add support for SPECspeed.
Add new os_types Centos7, Debian9, Ubuntu1404, Ubuntu1604, and Ubuntu1710.
Make it easier to RDP to PKB VMs
Add os_type support to KubernetesVirtualMachine.
Avoid setting up thread pool etc when run_processes is not set (to simplify
debugging).
Added a sample benchmark for descriptive purposes.
Added GPU peer to peer topology information to metadata.
Added a flag, mpirun_allow_run_as_root which allows OpenMPI to run in the
case that the user is root.
Modified KubernetesVirtualMachine to ensure that ssh is installed on the
container.
Added container_cluster_num_vms flag.
Added the cluster's worker shape and worker count into the dpb service metadata.
Added azure_accelerated_networking flag for Azure SR-IOV support.
Added support for V100 GPUs on GCP.
Add support for distributed TensorFlow benchmark.
Added Azure Relational DB support for Postgres 9.6.
Added copy_benchmark_single_file_mb flag for single file support.
Record guest system information.
Support for static edw clusters.
Add more granularity to FAILED benchmarks with FailedSubstatus due to
insufficient quota, cloud capacity, or known intermittent failures.
Update sysbench benchmark to version 1.0. (deprecate 0.4 and 0.5 versions)
Change GCP TPU command from alpha to beta.
Update configurable parameters for ycsb benchmarks.
Added tags to AWS kops instances and increased loadbalancer timeout.
Allow cassandra_ycsb to run on a single vm.
Adding support for using preprovisioned data for SPEC CPU 2006 benchmark.
Support negative numbers when parsing an integerlist.
Added float16 support for TensorFlow and MXNet Benchmarks.
Added flag --mx_key_value_store in MNXnet benchmark.
Added time_commands flag to enable diagnostic timing of commands
Added image processing speed in mnist.
Updated cloud TPU model link.
Updated AWS spot instance creation.
Added support for extending the failure sample with metadata if AWS spot VMs
or GCP preemptible VMs are interrupted.
Added flags ycsb_version and ycsb_measurement_type to support
user-specified ycsb versions and measurement types.
Added support to tensorflow_benchmark for running multiple batch sizes per run.
Added resnet152 in TensorFlow benchmark default models.
Added 50kb per payload ycsb workload.
Added num_cpus to virtual_machine published metadata.
Added a timeout to RobustRemoteCommand.
Added support for the gcp_min_cpu_platform flag on GKE clusters.
Preliminary support for NFS file systems
Added BigQuery provider implementation to extend the edw benchmarking in Perfkitbenchmarker.
Added support for authorizing gcloud to access Cloud Platform with a Google service account.
Added support for specification of resource specific ready timeouts.
Adding runner utilities to execute multiple sql scripts on edw clusters.
TPU checkpoints and summaries are now stored in GCS.
Updated cloud TPU models version.
Support for cloud NFS services (no implementation).
Added support for default batch sizes given a certain GPU type and model in
the Tensorflow benchmark.
Added method to get the NfsService from the linux_virtual_machine.
Added support for fio job files in the data directory.
Added InvalidConfigurationError.
Added owner tag in metadata.
Added support for NVIDIA P4 GPUs.
Added YCSB timeseries parsing.
Added Intel MKL to HPCC benchmark.
Added flag support for enabling or disabling transparent hugepages on Linux
VMs.
Add AWS MySql Aurora support
Abandon beta in TPU commands.
Update ycsb hdr histograms to output {bucket:count} data set for latencies, removing percentile:latency_value data set.
Added GPU test for ResNet benchmark.
Update ResNet testcases.
Remove unused lib in ResNet benchmark.
Added way to run several specsfs benchmarks in a single PKB run.
Add --preemptible in TPU creation command.
Add --skip_host_call in ResNet benchmark.
Update ResNet benchmark test for TensorFlow 1.9.
Removed the flag resnet_num_cores from the resnet benchmark and added the flag tpu_num_shards to MNIST, resnet, and inception3.
Update MNIST benchmark source path.
Replace old flags --tpu_name and --master with a new flag, --tpu.
Added support for EFS provisioned throughput.
Added support for --os_type ubuntu1804. Removed ubuntu1710 for GCP.
Added flag --iperf3_set_tcp_window_size to let the user avoid setting the
TCP Window size in Iperf3 Windows performance tests.
Added flags --hbase_version and --hbase_use_stable to set the HBase
version. Also upped hadoop to version 2.8.4.
Updated cuDNN installation methods.
Added support to schedule multiple TPUs.
Added GroupZonesIntoRegions support for providers.aws.util.
Added function to get the number of TPU cores.
Update spec17 copies per run to respect system available memory.
Add tensor2tensor benchmark.
Change train_steps to train_epochs in TPU test.
Add default tags like timeout_utc to GCE.
Add validation to all the TPU tests.
Add number of Train/Eval TPU shards in metadata.
The spark service resource supports a new flag (spark_service_log_level), to control the log level and generated output.
Updated openjdk 8 installation to support newer ycsb versions (after 0.13.0);
Improved hdrhistogram result reporting in ycsb.py.
Added flag --ntttcp_udp to allow the user to run UDP tests with ntttcp.
Added flag --ntttcp_packet_size to allow user to set the packet size in
ntttcp tests.
Extract more data from the ntttcp results file and put into the metadata of
the samples.
Updated the default of Cloud Bigtable client version to 1.4.0. Added --hbase_bin_url to allow bypassing GetHBaseURL().
Remove flag tf_benchmarks_commit_hash, and add tf_cnn_benchmarks_branch.
Branch cnn_tf_vX.Y_compatible is compatible with TensorFlow version X.Y.
Added flags --ntttcp_sender_rb, --ntttcp_sender_sb, --ntttcp_receiver_rb, --ntttcp_receiver_sb to control the socket buffer
sizes in ntttcp tests.
Move the ycsb_operation_count default from being set in the flag to being set in the workload file.
Introducing a new flag gcp_dataproc_subnet which allows a user to specify the subnet that a dataproc cluster will be part of.
Added support for running a subset of HPCC benchmarks.
Introducing a new multi string flag 'gcp_dataproc_property' which allows a user to modify many common configuration files when creating a Dataproc cluster.
Added configuration support for specifying type of the boot disk for a Dataproc cluster.
Added AddTag method to AzureResourceGroup.
Added some support for IPv6 (on static machines).
Added retransmit count to netperf metadata.
Added a common base class to PKB, PkbCommonTestCase,
which uses absl.testing.flagsaver to save and restore
flag values on test setUp / tearDown.
Remove all usage of mock_flags and replace with PkbCommonTestCase.
Upgraded memtier benchmark to version 1.2.15.
Add precision flag in Inception3 benchmark.
Support Hadoop 3.x.x in hadoop_terasort
Add z1 support for NVME disks.
Add default tags to dataproc.
Adding default run configurations for spec17 when none
of os_type, runspec_config or is_partial_results flags is set.
Add a flag to control the number of worker threads used by the
memcached server
Add default tags to GKE.
Add support for using real data from GCS and add support to download GCS data
to vm in the Tensorflow benchmark.
Added flag nttcp_config_list to allow the user to supply a list of test
configurations, all to be run in a single run phase.
Add support for --nouse_pkb_logging to use standard ABSL logging instead.
Improved support for booting more than 200 VMs with the cluster_boot benchmark.
Adding version support to redis server and setting permissions for newer redis versions.
Introduced app service metadata to indicate backend concurrency.
Added --ssh_reuse_connections to reuse SSH connections which speeds up benchmarks with lots of short commands.
Add abstract PreDelete method to the base resource class.
Add ability for linux vms to print the dmesg command before termination
using the flag '--log_dmesg'.
TPU and its GCS bucket should be in the same region.
Adding field count and field length flag overrides to ycsb load.
Added --ssh_server_alive_interval and ssh_server_alive_count_max to adjust SSH server alive options.
Support for cloud SMB services (no implementation).
Added rwmixread option to fio scenarios. This option is ignored unless a split
read/write workload is specified.
Added gce_local_ssd_count and gce_local_ssd_interface metadata to
GceVirtualMachine.
Added option to use real training data to the Horovod benchmark.
Added support to record individual latency samples in the object_storage_service
benchmark using the flag --record_individual_latency_samples.
Added --ssh_retries to adjust the number of times we retry for errors with a
255 error code.
Added --num_benchmark_copies which controls the number of copies of each
configuration to run.
Added --zone multistring flag as an additional way to specify zones. It can be
used in conjunction with the --zones flag.
Added --before_cleanup_pause to ease debugging.
Added support for CUDA 10.1.
Added --skip_firewall_rules flag to help manage firewall rule economy.
Added Azure Premium Files support.
Added 'os_type' metadata to virtual machines for use in performance samples.
Storage specification related dpb terasort benchmark improvements.
Added support to run fio against multiple clients.
Added an optional flag to set a delay between boot tasks.
Added precision flag to Horovod.
Added a flag to Horovod that indicates that the VM is using a deep learning
image.
Add support to set load upper bound for specsfs2014.
Add support for different workload sizes for stress-ng.
Implemented RobustRemoteCommand for Windows.
Extract GceVirtualMachine's GetNetwork into a method.
Updated redis memtier benchmark to pre-populate the redis db before testing.
Add Clear Linux support for AWS stripe disks.
Load memcached server before running benchmark.
Add support for xfs disk formatting.
Add gluster_fio benchmark.
Added support for static AWS VPCs with --aws_vpc and --aws_subnet flags.
Added support to append a region name to a bucket name for object storage
benchmarking. For Azure, the region is appended to storage accounts and
resource groups as well.
Added sysbench benchmarking for MySQL in a VM.
Added check that cuda_toolkit is installed when installing cudnn.
Added ability to set provider-specific MySQL flags for relational databases.
Add support for running t3 burstable VMs on AWS without unlimited mode.
Add support to run fio with a timeout on fio commands.
Added support for reporting bigtable cluster cpu utilization, gated by flag
--get_bigtable_cluster_cpu_utilization.
Move vm_groups to inside the relational_db spec when there is one, to allow it
to control what vms are used. Rename the old vm_spec and disk_spec under the
relational_db to db_spec and db_disk_spec respectively, for added clarity.
Added ability in stress_ng benchmark to test --cpu-method's.
Updated cloud_spanner_ycsb_benchmark to add the support for GCP Go client
library.
Make Kubernetes virtual machines not rebootable by default, to avoid changing
things that require rebooting, since doing so is not ideal for containers.
Add Amd Blis support for HPCC.
Add Stress-ng support to run on different versions, 0.05.23 and 0.09.25, which
are the defaults in Ubuntu1604 and Ubuntu1804.
Improved the usability of cloud bigtable ycsb benchmark by introducing flags
--google_bigtable_static_table_name,
--google_bigtable_enable_table_object_sharing, and --ycsb_skip_load_stage.
Added --cluster_boot_test_port_listening for cluster_boot benchmark. When
set to True it will report the time until the remote command port (SSH
port for Linux and WinRM port for Windows) in addition to the time until
a remote command is successfully run.
Added flag --cluster_boot_test_rdp_port_listening to measure time until RDP
port accepts a connection in Windows. It is True by default.
Add support for generating CLI download commands for object_storage_service.
Added flag '--azure_low_priority_vms' to create Low Priority VMs in Azure.
Bug fixes and maintenance updates:
Moved GPU-related specs from GceVmSpec to BaseVmSpec
Fix ProfitBricks issue with extra / in the API url
Fix ProfitBricks volume availability zone issue
Bulk AllowPort restored.
Moved CustomMachineTypeSpec and related decoders to their own module
Updated GKE engine version to 1.9.6-gke.0 when using a GPU-accelerated cluster
Update python and pip package names used when installed with Yum.
Don't try to publish samples if there aren't any
Disallow overwriting of benchmarks using the same benchmark name.
Log exception thrown in RunBenchmark
Explicitly state python package names for RedHat on GCP
Always set password for 'postgres' user when creating PostgreSQL CloudSQL
instances.
Fixed EPEL package URL for Redhat 7.
Updated to latest version of NVIDIA CUDA Toolkit 8.
Updated Ubuntu image versions in Tensorflow benchmark.
Don't log wget output when installing CUDA Toolkit 8.
Install CUDA Toolkit 8 patch from NVIDIA.
Updated Ubuntu image versions in MNIST and MXNet benchmarks.
Allow SMB3 protocol for file transfer which is required for windows 2012 with encrypted transport
Updated Ubuntu image versions in HPCG, MNIST and MXNet benchmarks.
Fixed a bug getting the correct number of layers in MXNet.
Gcloud alpha version no longer required for GPU and minimum CPU platform.
Backfill image on GCE if empty.
Remove logging of empty environment variables in vm_util.IssueCommand.
Explicitly state python package names for RedHat and Centos in AWS and Azure.
Remove flag "tf_use_nccl" because the latest TensorFlow benchmark removed this
argument.
"nightly" now is not a valid argument in GCP TPU commands.
Make aerospike_ycsb runnable on Amazon AMI.
Fixed assumption that HOME was set.
Fixed issue with default K8s image setup.
Install the correct TensorFlow pip package based on whether VM has GPUs or
not.
Change to root for command "lspci".
Install package pciutils when install cuda_toolkit.
Update TensorFlow benchmark arguments.
Isolate Darwin OS pkb runs to fix Issue # 1563
Fixed YCSB so it pushes workload once to each VM even if they are repeated.
Simplified resource and spec registration.
Fixed a bug in which the PKB recorded reboot time could be low for AWS because
we would restart the creation timer for VMs that were reported as pending.
Increased timeout of GKE Cluster create command from 10 to 15 minutes.
Set ImageNet image shape as 3,224,224.
Updated MountDisk commands to use named parameters.
Calling repr(IntegerList) returns a readable string.
Support installing azure-cli on RedHat systems.
Fixed default behavior of using /usr/bin/time --quiet on all commands
Fixed ycsb failure when the same workload is ran more than once.
Fixed yum proxy config bug (GH-#1598 from @kopecpiotr)
Create destination directory for Azure blob storage downloads.
Made the pip package more robust by adding a symbolic link in /usr/bin if pip
is installed in /usr/local/bin.
Make ~/.ssh/config only readable by the owner.
Increased timeout for Azure az vm create commands.
Increased timeout for GCP gcloud compute instances create commands.
Replace all underscores in the benchmark name with dashes when downloading
preprovisioned benchmark data from Azure. This is because Azure blob storage
container names cannot contain underscores.
Fixed the bug of running distributed TensorFlow on multiple batch sizes.
Updated gcloud compute networks create to use subnet-mode.
Changed default CUDA version from 8.0 to 9.0.
Updated default Tensorflow version to 1.7 when using GPUs.
Updated default Tensorflow tf_cnn_benchmarks version to a commit dated April
2, 2018.
Fixed a bug in which boto could not be uninstalled on newer GCE images.
Update each ycsb workload to have its own recordcount configuration if
ycsb_record_count is not set.
Fixed treatment of the boot time metric so that it is more like any other run
stage metric.
Fixed bug of modifying the providers/aws/util.AWS_PREFIX value.
Made failures to create VMs on AWS, Azure, and GCP fail PKB quickly.
Fix Kubernetes StorageClass deletion
Added git installation to tensorflow_serving package.
MountDisk throws exception if mounting the disk fails.
Added support for preprovisioned benchmark data on KubernetesVirtualMachines.
Refactored speccpu2006 benchmark to use common elements for both
speccpu2006 and speccpu2017.
Use flags['key'].parse(...) to set mocked flags in linux_virtual_machine_test
Cleanup some redundant logging and duplicate decoder statements.
Fixed build_tools package re-installation for speccpu2006.
Fixed fio job parsing, section parameters should always overrides global
job parameters.
Refactored StaticVirtualMachines to use GetResourceClass() pattern.
Fixing the Redshift provider implementation to extend the edw benchmarking in pkb.
Support using --gcp_min_cpu_platform=none to clear --min-cpu-platform. This
lets the flag override a benchmark spec.
Fix windows security protocol casting.
Added '--iteration' to MNIST benchmark. The default value of 50 is too small
and can result in excessive communication overhead which negatively impacts
performance.
Added support for more AWS regions in the object storage benchmark.
Add ability to skip known failing scripts when running edw profiles.
Set the control port and data port for nuttcp benchmark.
Fix overwriting of bandwidth variable in nuttcp benchmark.
Fixed fio histogram parsing.
Refactored AwsKeyFileManager out of AwsVirtualMachine.
Added delay between server and client initiation to windows benchmarks.
Defaulted static machines to linux based.
Add time limit to windows fio benchmark.
Adding -w buffer parameter to windows iperf3 benchmark.
Fixed a bug in how we select images on AWS by introducing an additional
regular expression to match against image names.
Terminate long running servers on windows benchmarks with timeouts.
Updated azure_cli package to match installation instructions.
Fix helpmatch for intergerlist.
Fix PSPing benchmark so that it runs on AWS and Azure.
Upgrade CPU pip package version in the Tensorflow benchmark to version 1.6.
Moved from ACS to AKS for Azure Kubernetes clusters.
Cleanup and fix Beam benchmark.
Sysbench failover tests added for GCP and AWS Aurora
Sysbench qps metric added
Fixed a bug of checking TPU exist.
Add GetMasterGrpcAddress method to CloudTpu.
Fix a bug of getting the number of TPU cores.
Set number of images in ResNet benchmark command so it can support other datasets.
Sysbench supports benchmarking MySQL 5.6
Update memcached server to install from a pre-built package provided by the operating system.
TensorFlow Serving benchmark now runs off master branch with optimized binaries
Updated HPCC benchmark to version 1.5.0.
Psping benchmark no longer report histogram for every sample metadata.
Specifies the number of threads to use in MXNet CPU test.
Fixed the way several unittests are using flags and mock_flags.
Fixed linter issues in several unittests.
Add python_pip_package_version as a class attribute of
BaseVirtualMachine. Children classes that have the problem of pypa/pip#5247 can use an older pip package.
Update AWS Aurora Postgres default version from 9.6.2 to 9.6.8
Update speccpu17 to write profiles in different files for each benchmark.
Avoid installing unnecessary MySQL server for sysbench when client tools suffice.
Added support for running the object service benchmark with a GCP service
account.
Add missing if name == 'main' stanza to some unittests, and fix those
that were broken.
Consolidate FlagDictSubstitution and FlagsDecoder into a simpler context manager,
OverrideFlags, which does not abuse the internals of FlagValues.
Changed GkeCluster so that it adds PKB metadata to all VMs that it creates.
Upgraded OpenMPI from 1.6.5 to 3.1.2.
Upgraded OpenBLAS from 0.2.15 to 0.3.3.
Add flag to control database machine type for managed relational databases.
Ensure randomly generated windows passwords start with a character.
Upgrade Tensorflow version to 1.12.
Install NCCL when installing Tensorflow with GPU support.
Update AzureBlobStorageService to persist for max of --timeout_minutes and --persistent_timeout_minutes.
Update S3Service to persist for max of --timeout_minutes and --persistent_timeout_minutes.
Add --project flag to GoogleCloudStorageService MakeBucket command.
Fix virtual_machine.GetResourceMetadata() so that it does not try to gather
metadata from a VM that has not been provisioned yet.
TPU Pod doesn't support eval. Stop Evaluation in train phrase in MNIST benchmark.
Global steps was set incorrectly in Inception3 benchmark.
Add AWS T3 instances to list of non-placement group capable machine types.
Refactor managed redis resource base class into managed memory store resource base class
Remove flag tpu_zone because VM and TPU should be in the same zone.
Refactoring managed_memorystore to add GetIp, GetPort and GetPassword class methods for managed_memorystore base class.
Run individual stressors in stress-ng one by one.
Upgrade gcp cloud redis to use prod gcloud.
Upgrade glibc_benchmark's version of glibc to 2.29.
Add number of disks for AWS i3.metal instance to configuration
Fix bug in Azure disk letter assignment when attaching more than 24 disks.
Incremental fixes to support Python3.
Fix bug in Azure Windows VM names for long run URIs where the name could be
longer than 15 characters.
Update aws dynamodb_ycsb_benchmark to use aws credentials from the runner vm.
Reboot immediately during call to ApplySysCtlPersient.
Set lower tcp keepalive thresholds on netperf vms.
Introduced NumCpusForBenchmark() which should be used instead of num_cpus for benchmark configuration.
Use stat -c %z /proc/ instead of uptime -s to find a boot timestamp since it
is more universal.
Added support for regex_util.ExtractInt to extract an integer from a string.
Updated Coremark to v1.01 and improved its error handling and output parsing.
Added flag support to Coremark for choosing a parallelism method. Defaults to
the existing mechanism of pthreads.
Make a temporary copy on VM's disk for scp on SMB.
Change the GCE Windows VM boot disk type to PD-STANDARD by default to better
match the default for GCE Linux VMs.
Fixed azure credentials package to avoid pushing the entire .azure directory.
Update bigtable benchmark to make configurations configurable, and report them.
RobustRemoteCommand now retries on ssh connections refused.
Fixed a bug that was causing AWS capacity reservation creation to fail.
GceNetworkSpec's zone is equal to the VM's zone
Creates the bucket from PKB rather than from within the VM that PKB launches
in TPU test.
Sets TPU zone MLPerf benchmark.
Compile Stress-ng version 0.05.23 rather than installing its package.
Added --gce_firewall_rules_clean_all option to network resource: clean up
all the firewall rules that depend on the PKB-created network.
Fix AWS stripe disks assumption that the NVME index for the boot drive is always 0.
Increased default timeout for booting Windows VMs to 40 minutes.
Expose filesystem type and block size in vm metadata.
Fixed problem with AWS network creation in aws_dpb_emr, aws_nfs_service, and
elastic_container_service.
Create the client_vm of the relational_db attribute of BenchmarkSpec from the
default vm_group if there's no clients group. Fixes pgbench benchmark failure.
Fixed use of numpy.piecewise to be numpy > 1.13 compliant
Fixed the cluster boot benchmark config to use SSD boot disk types for GCP and
Azure, adding support for specifying a boot_disk_type in Azure.
Fixed the default path of the spark example jar on an EMR cluster.
Check if an AWS default route exists before creation.
Fixed the broken reference to ycsb.YCSB_TAR_URL in the
cloud_spanner_ycsb_benchmark. Without this, the benchmark is not runnable.
Fixed bug of query failing to find AWS internet gateway after Delete called.
Abort and log when container resources are exhausted.
Add a retry on gcloud commands when rate limited.
Check if an AWS firewall rule exists before creating one by querying AWS.
Add flag overrides for server and client vm groups inside relational_db.
Update requirements.txt for Python 3
Upgrade hadoop to version 3.2.1.
Install the netcat-openbsd and zlib.h debian package for aerospike_ycsb.
Updated linux_virtual_machine.py to use the single source of truth of
perfkitbenchmark_keyfile path for non-static VM, expanding the use case of
stage-wise test, e.g., do prepare on machine A, do run stage on machine B.
Added --scp_connect_timeout and --ssh_connect_timeout flag to replace the
hard-coded values at the call sites of vm_util.GetSshOptions.
Added support for Container Optimized OS in the GCP Provider. Support is
currently limited to the cluster_boot benchmark.
Unifying metadata creation on cluster boot.
Added support for CoreOS Container Linux in GCP, AWS, and Azure Providers.
Support is currently limited to the cluster_boot benchmark.
PkbCommonTestCase extends absltest.TestCase for py3 test backports.
Update tox python version.
Remove the remote file before pushing a local file to the remote place. This
can avoid permission issue as the pushed file only allows user to read or
execute.
