Benchmarking Lustre File System Performance (Lustre I/O Kit)

This chapter describes the Lustre I/O kit, a collection of I/O benchmarking tools for a Lustre cluster. It includes:

• Using Lustre I/O Kit Tools
• Testing I/O Performance of Raw Hardware (sgpdd-survey)
• Testing OST Performance (obdfilter-survey)
• Testing OST I/O Performance (ost-survey)
• Testing MDS Performance (mds-survey)

Using Lustre I/O Kit Tools

The tools in the Lustre I/O Kit are used to benchmark Lustre file system hardware and validate that it is working as expected before you install the Lustre software. It can also be used to to validate the performance of the various hardware and software layers in the cluster and also to find and troubleshoot I/O issues.

Typically, performance is measured starting with single raw devices and then proceeding to groups of devices. Once raw performance has been established, other software layers are then added incrementally and tested.

Contents of the Lustre I/O Kit

The I/O kit contains three tests, each of which tests a progressively higher layer in the Lustre software stack:

• sgpdd-survey - Measure basic ‘bare metal’ performance of devices while bypassing the kernel block device layers, buffer cache, and file system.
• obdfilter-survey - Measure the performance of one or more OSTs directly on the OSS node or alternately over the network from a Lustre client.
• ost-survey - Performs I/O against OSTs individually to allow performance comparisons to detect if an OST is performing sub-optimally due to hardware issues.

Typically with these tests, a Lustre file system should deliver 85-90% of the raw device performance.

Preparing to Use the Lustre I/O Kit

The following prerequisites must be met to use the tests in the Lustre I/O kit:

• Password-free remote access to nodes in the system (provided by ssh or rsh).
• LNet self-test completed to test that Lustre networking has been properly installed and configured. See Testing Lustre Network Performance (LNet Self-Test).
• Lustre file system software installed.
• sg3_utils package providing the sgp_dd tool (sg3_utils is a separate RPM package available online using YUM).

Download the Lustre I/O kit (lustre-iokit)from:

https://downloads.whamcloud.com/

Testing I/O Performance of Raw Hardware (sgpdd-survey)

The sgpdd-survey tool is used to test bare metal I/O performance of the raw hardware, while bypassing as much of the kernel as possible. This survey may be used to characterize the performance of a SCSI device by simulating an OST serving multiple stripe files. The data gathered by this survey can help set expectations for the performance of a Lustre OST using this device.

The script uses sgp_dd to carry out raw sequential disk I/O. It runs with variable numbers of sgp_dd threads to show how performance varies with different request queue depths.

The script spawns variable numbers of sgp_dd instances, each reading or writing a separate area of the disk to demonstrate performance variance within a number of concurrent stripe files.

Several tips and insights for disk performance measurement are described below. Some of this information is specific to RAID arrays and/or the Linux RAID implementation.

• Performance is limited by the slowest disk. Before creating a RAID array, benchmark all disks individually. We have frequently encountered situations where drive performance was not consistent for all devices in the array. Replace any disks that are significantly slower than the rest.
• Disks and arrays are very sensitive to request size. To identify the optimal request size for a given disk, benchmark the disk with different record sizes ranging from 4 KB to 1 to 2 MB.

Prerequisites:

• sgp_dd tool in the sg3_utils package
• Lustre software is NOT required

The device(s) being tested must meet one of these two requirements:

• If the device is a SCSI device, it must appear in the output of sg_map (make sure the kernel module sg is loaded).
• If the device is a raw device, it must appear in the output of raw -qa.

Raw and SCSI devices cannot be mixed in the test specification.

Tuning Linux Storage Devices

To get large I/O transfers (1 MB) to disk, it may be necessary to tune several kernel parameters as specified:

/sys/block/sdN/queue/max_sectors_kb = 4096
/sys/block/sdN/queue/max_phys_segments = 256
/proc/scsi/sg/allow_dio = 1
/sys/module/ib_srp/parameters/srp_sg_tablesize = 255
/sys/block/sdN/queue/scheduler

Running sgpdd-survey

The sgpdd-survey script must be customized for the particular device being tested and for the location where the script saves its working and result files (by specifying the ${rslt} variable). Customization variables are described at the beginning of the script.

When the sgpdd-survey script runs, it creates a number of working files and a pair of result files. The names of all the files created start with the prefix defined in the variable ${rslt}. (The default value is /tmp. ) The files include:

• File containing standard output data (same as stdout)

  rslt_date_time.summary

• Temporary (tmp) files

  rslt_date_time_*

• Collected tmp files for post-mortem

  rslt_date_time.detail

The stdout and the .summary file will contain lines like this:

total_size 8388608K rsz 1024 thr 1 crg 1 180.45 MB/s 1 x 180.50 \
        = 180.50 MB/s

Each line corresponds to a run of the test. Each test run will have a different number of threads, record size, or number of regions.

• total_size - Size of file being tested in KBs (8 GB in above example).
• rsz - Record size in KBs (1 MB in above example).
• thr - Number of threads generating I/O (1 thread in above example).
• crg - Current regions, the number of disjoint areas on the disk to which I/O is being sent (1 region in above example, indicating that no seeking is done).
• MB/s - Aggregate bandwidth measured by dividing the total amount of data by the elapsed time (180.45 MB/s in the above example).
• MB/s - The remaining numbers show the number of regions X performance of the slowest disk as a sanity check on the aggregate bandwidth.

If there are so many threads that the sgp_dd script is unlikely to be able to allocate I/O buffers, then ENOMEM is printed in place of the aggregate bandwidth result.

If one or more sgp_dd instances do not successfully report a bandwidth number, then FAILED is printed in place of the aggregate bandwidth result.

Testing OST Performance (obdfilter-survey)

The obdfilter-survey script generates sequential I/O from varying numbers of threads and objects (files) to simulate the I/O patterns of a Lustre client.

The obdfilter-survey script can be run directly on the OSS node to measure the OST storage performance without any intervening network, or it can be run remotely on a Lustre client to measure the OST performance including network overhead.

The obdfilter-survey is used to characterize the performance of the following:

• Local file system - In this mode, the obdfilter-survey script exercises one or more instances of the obdfilter directly. The script may run on one or more OSS nodes, for example, when the OSSs are all attached to the same multi-ported disk subsystem. Run the script using the case=disk parameter to run the test against all the local OSTs. The script automatically detects all local OSTs and includes them in the survey. To run the test against only specific OSTs, run the script using the targets=parameter to list the OSTs to be tested explicitly. If some OSTs are on remote nodes, specify their hostnames in addition to the OST name (for example, oss2:lustre-OST0004). All obdfilter instances are driven directly. The script automatically loads the obdecho module (if required) and creates one instance of echo_client for each obdfilter instance in order to generate I/O requests directly to the OST. For more details, see Testing Local Disk Performance.
• Network - In this mode, the Lustre client generates I/O requests over the network but these requests are not sent to the OST file system. The OSS node runs the obdecho server to receive the requests but discards them before they are sent to the disk. Pass the parameters case=network and targets=*hostname|IP_of_server* to the script. For each network case, the script does the required setup. For more details, see Testing Network Performance
• Remote file system over the network - In this mode the obdfilter-survey script generates I/O from a Lustre client to a remote OSS to write the data to the file system. To run the test against all the local OSCs, pass the parameter case=netdisk to the script. Alternately you can pass the target= parameter with one or more OSC devices (e.g., lustre-OST0000-osc-ffff88007754bc00) against which the tests are to be run. For more details, see Testing Remote Disk Performance.

Testing Local Disk Performance

The obdfilter-survey script can be run automatically or manually against a local disk. This script profiles the overall throughput of storage hardware, including the file system and RAID layers managing the storage, by sending workloads to the OSTs that vary in thread count, object count, and I/O size.

When the obdfilter-survey script is run, it provides information about the performance abilities of the storage hardware and shows the saturation points.

The plot-obdfilter script generates from the output of the obdfilter-survey a CSV file and parameters for importing into a spreadsheet or gnuplot to visualize the data.

To run the obdfilter-survey script, create a standard Lustre file system configuration; no special setup is needed.

To perform an automatic run:

1. Start the Lustre OSTs. The Lustre OSTs should be mounted on the OSS node(s) to be tested. The Lustre client is not required to be mounted at this time.
2. Verify that the obdecho module is loaded. Run:

modprobe obdecho

3. Run the obdfilter-survey script with the parameter case=disk. For example, to run a local test with up to two objects (nobjhi), up to two threads (thrhi), and 1024 MB transfer size (size):

$ nobjhi=2 thrhi=2 size=1024 case=disk sh obdfilter-survey

4. Performance measurements for write, rewrite, read etc are provided below:

# example output
Fri Sep 25 11:14:03 EDT 2015 Obdfilter-survey for case=disk from hds1fnb6123
ost 10 sz 167772160K rsz 1024K obj   10 thr   10 write 10982.73 [ 601.97,2912.91] rewrite 15696.54 [1160.92,3450.85] read 12358.60 [ 938.96,2634.87]
...

The file ./lustre-iokit/obdfilter-survey/README.obdfilter-survey provides an explaination for the output as follows:

ost 10          is the total number of OSTs under test.
sz 167772160K   is the total amount of data read or written (in bytes).
rsz 1024K       is the record size (size of each echo_client I/O, in bytes).
obj    10       is the total number of objects over all OSTs
thr    10       is the total number of threads over all OSTs and objects
write           is the test name.  If more tests have been specified they
           all appear on the same line.
10982.73        is the aggregate bandwidth over all OSTs measured by
           dividing the total number of MB by the elapsed time.
[601.97,2912.91] are the minimum and maximum instantaneous bandwidths seen on
           any individual OST.
Note that although the numbers of threads and objects are specifed per-OST
in the customization section of the script, results are reported aggregated
over all OSTs.

To perform a manual run:

1. Start the Lustre OSTs. The Lustre OSTs should be mounted on the OSS node(s) to be tested. The Lustre client is not required to be mounted at this time.
2. Verify that the obdecho module is loaded. Run:

modprobe obdecho

3. Determine the OST names. On the OSS nodes to be tested, run the lctl dl command. The OST device names are listed in the fourth column of the output. For example:

$ lctl dl |grep obdfilter
0 UP obdfilter lustre-OST0001 lustre-OST0001_UUID 1159
2 UP obdfilter lustre-OST0002 lustre-OST0002_UUID 1159
...

4. List all OSTs you want to test. Use the targets=parameter to list the OSTs separated by spaces. List the individual OSTs by name using the format *fsname*-*OSTnumber* (for example, lustre-OST0001). You do not have to specify an MDS or LOV.
5. Run the obdfilter-survey script with the targets=parameter. For example, to run a local test with up to two objects (nobjhi), up to two threads ( thrhi), and 1024 Mb (size) transfer size:

$ nobjhi=2 thrhi=2 size=1024 targets="lustre-OST0001 \
          lustre-OST0002" sh obdfilter-survey

Testing Network Performance

The obdfilter-survey script can only be run automatically against a network; no manual test is provided.

To run the network test, a specific Lustre file system setup is needed. Make sure that these configuration requirements have been met.

To perform an automatic run:

1. Start the Lustre OSTs. The Lustre OSTs should be mounted on the OSS node(s) to be tested. The Lustre client is not required to be mounted at this time.
2. Verify that the obdecho module is loaded. Run:

modprobe obdecho

3. Start lctl and check the device list, which must be empty. Run:

lctl dl

4. Run the obdfilter-survey script with the parameters case=network and targets=*hostname|ip_of_server* . For example:

$ nobjhi=2 thrhi=2 size=1024 targets="oss0 oss1" \
            case=network sh obdfilter-survey

5. On the server side, view the statistics at:

lctl get_param obdecho.echo_srv.stats

where *echo_srv* is the obdecho server created by the script.

Testing Remote Disk Performance

The obdfilter-survey script can be run automatically or manually against a network disk. To run the network disk test, start with a standard Lustre configuration. No special setup is needed.

To perform an automatic run:

1. Start the Lustre OSTs. The Lustre OSTs should be mounted on the OSS node(s) to be tested. The Lustre client is not required to be mounted at this time.
2. Verify that the obdecho module is loaded. Run:

modprobe obdecho

3. Run the obdfilter-survey script with the parameter case=netdisk. For example:

$ nobjhi=2 thrhi=2 size=1024 case=netdisk sh obdfilter-survey

To perform a manual run:

1. Start the Lustre OSTs. The Lustre OSTs should be mounted on the OSS node(s) to be tested. The Lustre client is not required to be mounted at this time.
2. Verify that the obdecho module is loaded. Run: modprobe obdecho
3. Determine the OSC names. On the OSS nodes to be tested, run the lctl dl command. The OSC device names are listed in the fourth column of the output. For example:

$ lctl dl |grep obdfilter
3 UP osc lustre-OST0000-osc-ffff88007754bc00 \
           54b91eab-0ea9-1516-b571-5e6df349592e 5
4 UP osc lustre-OST0001-osc-ffff88007754bc00 \
           54b91eab-0ea9-1516-b571-5e6df349592e 5
...

4. List all OSCs you want to test. Use the targets=parameter to list the OSCs separated by spaces. List the individual OSCs by name separated by spaces using the format *fsname*-*OST_name*-osc-*instance* (for example, lustre-OST0000-osc-ffff88007754bc00). You do not have to specify an MDS or LOV.
5. Run the obdfilter-survey script with the targets=*osc* and case=netdisk. An example of a local test run with up to two objects (nobjhi), up to two threads (thrhi), and 1024 Mb (size) transfer size is shown below:

$ nobjhi=2 thrhi=2 size=1024 \
targets="lustre-OST0000-osc-ffff88007754bc00 \
lustre-OST0001-osc-ffff88007754bc00" sh obdfilter-survey

Output Files

When the obdfilter-survey script runs, it creates a number of working files and a pair of result files. All files start with the prefix defined in the variable ${rslt}.

The obdfilter-survey script iterates over the given number of threads and objects performing the specified tests and checks that all test processes have completed successfully.

Script Output

The .summary file and stdout of the obdfilter-survey script contain lines like:

ost 8 sz 67108864K rsz 1024 obj 8 thr 8 write 613.54 [ 64.00, 82.00]

Where:

Visualizing Results

It is useful to import the obdfilter-survey script summary data (it is fixed width) into Excel (or any graphing package) and graph the bandwidth versus the number of threads for varying numbers of concurrent regions. This shows how the OSS performs for a given number of concurrently-accessed objects (files) with varying numbers of I/Os in flight.

It is also useful to monitor and record average disk I/O sizes during each test using the ‘disk io size’ histogram in the file lctl get_param obdfilter.*.brw_stats (see Monitoring the OST Block I/O Stream for details). These numbers help identify problems in the system when full-sized I/Os are not submitted to the underlying disk. This may be caused by problems in the device driver or Linux block layer.

The plot-obdfilter script included in the I/O toolkit is an example of processing output files to a .csv format and plotting a graph using gnuplot.

Testing OST I/O Performance (ost-survey)

The ost-survey tool is a shell script that uses lfs setstripe to perform I/O against a single OST. The script writes a file (currently using dd) to each OST in the Lustre file system, and compares read and write speeds. The ost-survey tool is used to detect anomalies between otherwise identical disk subsystems.

To run the ost-survey script, supply a file size (in KB) and the Lustre file system mount point. For example, run:

$ ./ost-survey.sh -s 10 /mnt/lustre

Typical output is:

Number of Active OST devices : 4
Worst  Read OST indx: 2 speed: 2835.272725
Best   Read OST indx: 3 speed: 2872.889668
Read Average: 2852.508999 +/- 16.444792 MB/s
Worst  Write OST indx: 3 speed: 17.705545
Best   Write OST indx: 2 speed: 128.172576
Write Average: 95.437735 +/- 45.518117 MB/s
Ost#  Read(MB/s)  Write(MB/s)  Read-time  Write-time
----------------------------------------------------
0     2837.440       126.918        0.035      0.788
1     2864.433       108.954        0.035      0.918
2     2835.273       128.173        0.035      0.780
3     2872.890       17.706        0.035      5.648

Testing MDS Performance (mds-survey)

The mds-survey script tests the local metadata performance using the echo_client to drive the MDD layer of the MDS stack. It can be used with the following classes of operations:

• Open-create/mkdir/create
• Lookup/getattr/setxattr
• Delete/destroy
• Unlink/rmdir

These operations will be run by a variable number of concurrent threads and will test with the number of directories specified by the user. The run can be executed such that all threads operate in a single directory (dir_count=1) or in private/unique directory (dir_count=x thrlo=x thrhi=x).

The mdd instance is driven directly. The script automatically loads the obdecho module if required and creates instance of echo_client.

This script can also create OST objects by providing stripe_count greater than zero.

To perform a run:

1. Start the Lustre MDT. The Lustre MDT should be mounted on the MDS node to be tested.
2. Start the Lustre OSTs (optional, only required when test with OST objects) The Lustre OSTs should be mounted on the OSS node(s).
3. Run the mds-survey script as explain below The script must be customized according to the components under test and where it should keep its working files. Customization variables are described as followed:

• thrlo - threads to start testing. skipped if less than dir_count
• thrhi - maximum number of threads to test
• targets - MDT instance
• file_count - number of files per thread to test
• dir_count - total number of directories to test. Must be less than or equal to thrhi
• stripe_count - number stripe on OST objects
• tests_str - test operations. Must have at least “create” and “destroy”
• start_number - base number for each thread to prevent name collisions
• layer - MDS stack’s layer to be tested Run without OST objects creation: Setup the Lustre MDS without OST mounted. Then invoke the mds-survey script

$ thrhi=64 file_count=200000 sh mds-survey

Run with OST objects creation: Setup the Lustre MDS with at least one OST mounted. Then invoke the mds-survey script with stripe_count parameter

$ thrhi=64 file_count=200000 stripe_count=2 sh mds-survey

Note: a specific MDT instance can be specified using targets variable.

$ targets=lustre-MDT0000 thrhi=64 file_count=200000 stripe_count=2 sh mds-survey

Output Files

When the mds-survey script runs, it creates a number of working files and a pair of result files. All files start with the prefix defined in the variable ${rslt}.

The mds-survey script iterates over the given number of threads performing the specified tests and checks that all test processes have completed successfully.

Script Output

The .summary file and stdout of the mds-survey script contain lines like:

mdt 1 file 100000 dir 4 thr 4 create 5652.05 [ 999.01,46940.48] destroy 5797.79 [ 0.00,52951.55]

Where: