Skip to content

1. Getting Started

lukemartinlogan edited this page May 16, 2023 · 27 revisions

Building Hermes

There are several ways to obtain a working Hermes installation. Information on dependencies can be found in the README.

  1. Docker Image
  1. CMake
  • Instructions can be found in the README
  1. Spack
  • Instructions can be found in the README

If you get stuck, the root of the repository contains a ci folder where we keep the scripts we use to build and test Hermes in a Github Actions workflow. The workflow file itself is here.

Deploying Resources

Hermes is an application extension. Storage resources are deployed under Hermes control by

  1. Configuring Hermes for your system and application
  2. Making your application "Hermes-aware"

An application can be made aware of Hermes in at least three different ways:

  • Through Hermes adapters, LD_PRELOAD-able shared libraries which intercept common I/O middleware calls such as UNIX STDIO, POSIX, and MPI-IO (NOTE: when Hermes is compiled with DHERMES_USE_ADDRESS_SANITIZER=ON, which is ON by default, you must also ensure that libasan is preloaded first, before anything else)
  • Through an HDF5 virtual file driver (VFD)
  • By directly targeting the Hermes native API

These options represent different use cases and trade-offs, for example, with respect to expected performance gains and required code change.

Adapters

When using the STDIO adapter (intercepting fopen, fwrite, etc.) and the POSIX adapter (intercepting open, write, etc.), there are multiple ways to deploy Hermes with an existing application.

NOTE: The MPI-IO adapter is still experimental, and only supports MPICH at this time.

A note about Address Sanitization (libasan)

If you compile hermes with DHERMES_USE_ADDRESS_SANITIZER=ON, you must LD_PRELOAD the libasan used to build Hermes, in addition to the interceptor. To locate libasan, run the following command:

gcc -print-file-name=libasan.so

Note that libasan will detect memory leaks and errors in the program linking to hermes as well. To avoid detecting memory leaks in the client program, do the following:

# Create or modify a file for storing libasan exclusions:
nano ${HERMES_ROOT}/test/data/asan.supp

# Set the libasan environment variable to point to the file
LSAN_OPTIONS=suppressions=${HERMES_ROOT}/test/data/asan.supp

Check the example in the section below (Hermes services running in same process as the application) to see how to use hermes + asan together.

Deploying an Application with Hermes

The Hermes daemon is responsible for tracking various metadata, and it is required to be launched before your application. There should only be one Hermes daemon per node. This can be accomplished using SSH or MPI. The following example uses MPICH to deploy hermes on a cluster of two nodes. We then use MPICH to finalize the daemon and flush all remaining content back to the PFS.

# We must start one and only one Hermes daemon on each node.
# This job is started in the background so the terminal doesn't block forever
# HERMES_CONF is the configuration of the server.
mpirun -n 2 -ppn 1 \
  -genv PATH=${PATH} \
  -genv LD_LIBRARY_PATH=${LD_LIBRARY_PATH} \
  -genv HERMES_CONF=/path/to/hermes.yaml \
  ${HERMES_INSTALL_DIR}/bin/hermes_daemon &

# Now we can start our application
# HERMES_CONF is the same as the one when spawning the daemon
# HERMES_CLIENT_CONF contains any parameters relevant to the specific program
mpirun -n 4 -ppn 2 \
  -genv PATH=${PATH} \
  -genv LD_LIBRARY_PATH=${LD_LIBRARY_PATH} \
  -genv LD_PRELOAD=${HERMES_INSTALL_DIR}/lib/libhermes_posix.so \
  -genv HERMES_CLIENT_CONF=/path/to/hermes_client.yaml \
  -genv HERMES_CONF=/path/to/hermes_server.yaml \
  ./my_app

# Now we can finalize
# This will automatically flush all dirty data remaining back to the PFS
HERMES_CONF=/path/to/hermes.yaml \
${HERMES_INSTALL_DIR}/bin/finalize_hermes

Hermes Tutorial

Here we will walk through an entire example of using Hermes with IOR. IOR supports several I/O APIs (-a option), including POSIX, MPI-IO, and HDF5. Hermes has adapters for POSIX, MPI-IO, and HDF5. For serial (single process) HDF5, the Hermes VFD can be enabled via environment variable as described here. Parallel HDF5 can use the MPI-IO adapter. For this tutorial, we'll focus on POSIX. We assume you already have working Hermes and IOR installations. See the README for Hermes installation details.

Workload

We will simulate a checkpoint/restart workload in which a group of processes each write a checkpoint file, and then another group of processes on different nodes reads the checkpoint files. In the default case, the checkpoint files will be written to and then read from the parallel file system. When running with Hermes, the data will be buffered in fast, local media resulting in a nice speedup with no code changes required.

Target system

Clients

I'm running on a cluster with 8 client nodes, each with the following characteristics:

Intel(R) Xeon(R) Silver 4114 CPU @ 2.20GHz
40 cores (Hyperthreading enabled)
46 GiB DRAM
40 Gbps ethernet with RoCE capability

Storage Tiers

Name Description Measured Write Bandwidth
PFS OrangeFS running on 8 server nodes, backed by HDDs 536 MiB/s
NVMe Node-local NVMe attached SSDs. 1918 MiB/s
RAM Node-local DRAM. 79,061 MiB/s

Hermes configuration

Here we describe the Hermes configuration format. Hermes has two configurations: one for the daemon and one for the client program. We will briefly discuss each here. See Configuration for more details.

Daemon (server) configuration

For a documented example of how to create a Hermes configuration, please check the default configuration. Note, the default config is designed for single-node cases. We use YAML to define the Hermes configuration format.

Defining the buffering locations

First, we should define the kind of storage devices that are targeted for intermediate buffering.

devices:
  ram:
    mount_point: ""
    capacity: 50MB
    block_size: 4KB
    slab_sizes: [ 4KB, 16KB, 64KB, 1MB ]
    bandwidth: 6000MBps
    latency: 15us
    is_shared_device: false
    borg_capacity_thresh: [0.0, 1.0]
  nvme:
    mount_point: "/mnt/nvme/hermes_nvme"
    capacity: 100MB
    block_size: 4KB
    slab_sizes: [ 4KB, 16KB, 64KB, 1MB ]
    bandwidth: 1GBps
    latency: 600us
    is_shared_device: false
    borg_capacity_thresh: [ 0.0, 1.0 ]
  pfs:
    mount_point: "${HOME}/hermes_pfs"
    capacity: 100MB
    block_size: 64KB # The stripe size of PFS
    slab_sizes: [ 4KB, 16KB, 64KB, 1MB ]
    bandwidth: 100MBps # Per-device bandwidth
    latency: 200ms
    is_shared_device: true
    borg_capacity_thresh: [ 0.0, 1.0 ]

Here we have a YAML dictionary called devices. A semantic name is then provided for each device targeted for buffering. To tell Hermes a device is considered RAM, we use mount_point being the empty string. For other devices, this is can be a path to a directory located on a filesystem.

Running

IOR Baseline

mpirun -n 48 -ppn 12 \
  ior -w -r -o /PFS/USER/ior.out -t 1m -b 128m -F -e -Y -C -O summaryFormat=CSV

Here we launch 48 IOR processes across 4 nodes. The IOR options are explained in the following table.

Flag Description
-w Perform write
-r Perform read
-o Output/Input file
-t Size per write
-b Total I/O size per rank
-F Create one file for each process
-e Call fsync on file close.
-Y Call fsync after each write.
-C Shuffle ranks so that they read from different nodes than they wrote to
-O summaryFormat Show the output in a compact, CSV format

Some of these options require justification.

  • -Y: We do direct (non-buffered) I/O in order to simulate a situation with high RAM pressure. If the application is using most of the RAM, then the OS page cache will have less RAM available for buffering.
  • -C: This option simulates a situation where different nodes read the checkpoint than the ones that wrote it, resulting in a situation where the checkpoint cannot be read from the page cache, and forcing the app to go to the PFS.

Here are the results:

access,bw(MiB/s),IOPS,Latency,block(KiB),xfer(KiB),open(s),wr/rd(s),close(s),total(s),numTasks,iter
write,30.3120,30.3795,1.5543,131072.0000,1024.0000,15.0489,202.2413,35.0600,202.6921,48,0
read,2012.0224,2026.8185,0.0158,131072.0000,1024.0000,0.4558,3.0314,1.0307,3.0536,48,0

Our write bandwidth is 30 MiB/s and our read bandwidth is 2012 MiB/s.

IOR with Hermes

To enable Hermes with an IOR checkpoint/restart workload, we must start a daemon, LD_PRELOAD a Hermes adapter and set some environment variables.

NOTE: As a temporary workaround to issue #258 we must comment out the line backend->close(fd, params->backend_options); in ior.c:TestIoSys before compiling IOR. This change is implemented in the chogan/hermes branch of the IOR fork here.

We spawn a daemon on each node, then run our app with the appropriate environment variables, similar to the process described above.

HERMES_CONF_PATH=/absolute/path/to/hermes.yaml

# Start one daemon on each node
mpirun -n 4 -ppn 1 \
  -genv HERMES_CONF ${HERMES_CONF_PATH} \
  ${HERMES_INSTALL_DIR}/bin/hermes_daemon &

# Give the daemons a chance to initialize
sleep 3

# Start "checkpoint" app
mpirun -n 48 -ppn 12 \
  -genv LD_PRELOAD ${HERMES_INSTALL_DIR}/lib/libhermes_posix.so \
  -genv HERMES_CONF ${HERMES_CONF_PATH} \
  -genv HERMES_CLIENT 1 \
  -genv ADAPTER_MODE SCRATCH \
  -genv HERMES_STOP_DAEMON 0 \
  ior -w -k -o ${CHECKPOINT_FILE} -t 1m -b 128m -F -e -Y -O summaryFormat=CSV

# Start the "restart" app
mpirun -n 48 -ppn 8 \
  -genv LD_PRELOAD ${HERMES_INSTALL_DIR}/lib/libhermes_posix.so \
  -genv HERMES_CONF ${HERMES_CONF_PATH} \
  -genv HERMES_CLIENT 1 \
  -genv ADAPTER_MODE SCRATCH \
  ior -r -o ${CHECKPOINT_FILE} -t 1m -b 128m -F -e -O summaryFormat=CSV

Results:

access,bw(MiB/s),IOPS,Latency,block(KiB),xfer(KiB),open(s),wr/rd(s),close(s),total(s),numTasks,iter
write,1748.0946,1928.3122,0.0122,131072,1024,1.9385,3.1862,0.8167,3.5147
read,2580.2756,2861.2635,0.0074,131072,1024,0.1923,2.1473,1.7458,2.3811

We get a nice boost in write bandwidth, and a modest speedup in read bandwidth, all with no code changes.

We haven't done any performance optimization yet, so I expect to bridge the gap significantly between the 2.5 GiB read speed of Hermes and the baseline speed of reading from RAM (tmpfs on /dev/shm) with IOR of 60 GiB.