IOPS – techish.net

Performance tuning iSCSI Round Robin policies in ESXi for Nimble storage

Here’s an ESXi console script to loop through each Nimble eui.* adapter and set IOPS=0 and BYTES=0 (per Nimble recommendations).

for x in `esxcli storage nmp device list | awk '/Nimble iSCSI Disk/{print $7}' | sed -e 's/(//' -e 's/)//'`; do
echo $x
esxcli storage nmp psp roundrobin deviceconfig set -d $x -t bytes -B 0;
esxcli storage nmp psp roundrobin deviceconfig set -d $x -t iops -I 0 ;
esxcli storage nmp psp roundrobin deviceconfig get -d $x;
done

Note: If you change the order above and set bytes after iops, then the policy will be based on bytes and not IOPS.
To reset defaults, use the following script on the ESXi host console:

for x in `esxcli storage nmp device list | awk '/Nimble iSCSI Disk/{print $7}' | sed -e 's/(//' -e 's/)//'`; do
echo $x
esxcli storage nmp psp roundrobin deviceconfig set -d $x -t bytes -B 10485760;
esxcli storage nmp psp roundrobin deviceconfig set -d $x -t iops -I 1000 ;
esxcli storage nmp psp roundrobin deviceconfig set -d $x -t default;
esxcli storage nmp psp roundrobin deviceconfig get -d $x;
done

To make sure this survives a reboot, you can set a policy:

esxcli storage nmp satp rule add --psp=VMW_PSP_RR --satp=VMW_SATP_ALUA --vendor=Nimble --psp-option=policy=iops;iops=0

Note that if you previously configured a user-defined SATP rule for Nimble volumes to simply use the Round Robin PSP (per the Nimble VMware best practices guide), you will first need to remove that simpler rule, before you can add the above rule, or else you will get an error message that a duplicate user-defined rule exists. The command to remove the simpler rule is: –Bill

esxcli storage nmp satp rule remove --psp=VMW_PSP_RR --satp=VMW_SATP_ALUA --vendor=Nimble

My Notes and Benchmarks on VMware Flash Read Cache

I’ve spent some time exploring and studying the use and configuration of VMware Flash Read Cache (vFRC) and its benefits. These are my notes.
Useful Resources

On a guest virtual machine, vFRC is configured in Disk configuration area. The virtual machine needs to be on version 10 hardware. vSphere needs to be minimum version 5.5.

Benchmarks

I took a baseline benchmark of a simple Windows Server 2016 virtual machine that had a thin provisioned 20GB disk using DskSpd (formerly sqlio). The virtual machine disk disk is connected to an IBM DS3400 LUN with 4 x 300GB 15k RPM disks in RAID-10.
Baseline Virtual Machine

OS: Windows Server 2016
vCPU: 1
vRAM: 4GB
SCSI Controller: LSI Logic SAS
Virtual Disk: 40GB thin provisioned
Virtual machine hardware: vmx-10
Virtual Flash Read Cache: 0

Some notes before running a test. This is geared toward SQL workloads and identifies the type of I/O for the different SQL workload.

DskSpd test

diskspd.exe -c30G -d300 -r -w0 -t8 -o8 -b8K -h -L E:	estfile.dat

Results of testing.

Command Line: diskspd.exe -c30G -d300 -r -w0 -t8 -o8 -b8K -h -L E:	estfile.dat
Input parameters:
        timespan:   1
        -------------
        duration: 300s
        warm up time: 5s
        cool down time: 0s
        measuring latency
        random seed: 0
        path: 'E:	estfile.dat'
                think time: 0ms
                burst size: 0
                software cache disabled
                hardware write cache disabled, writethrough on
                performing read test
                block size: 8192
                using random I/O (alignment: 8192)
                number of outstanding I/O operations: 8
                thread stride size: 0
                threads per file: 8
                using I/O Completion Ports
                IO priority: normal
Results for timespan 1:
*******************************************************************************
actual test time:       301.18s
thread count:           8
proc count:             1
CPU |  Usage |  User  |  Kernel |  Idle
-------------------------------------------
   0|  99.23%|   7.05%|   92.18%|   0.77%
-------------------------------------------
avg.|  99.23%|   7.05%|   92.18%|   0.77%
Total IO
thread |       bytes     |     I/Os     |     MB/s   |  I/O per s |  AvgLat  | LatStdDev |  file
-----------------------------------------------------------------------------------------------------
     0 |      7940784128 |       969334 |      25.14 |    3218.43 |    2.471 |    22.884 | E:	estfile.dat (30GB)
     1 |      8152604672 |       995191 |      25.81 |    3304.28 |    2.401 |    22.211 | E:	estfile.dat (30GB)
     2 |      8116256768 |       990754 |      25.70 |    3289.55 |    2.408 |    22.080 | E:	estfile.dat (30GB)
     3 |      8180006912 |       998536 |      25.90 |    3315.38 |    2.394 |    22.936 | E:	estfile.dat (30GB)
     4 |      8192147456 |      1000018 |      25.94 |    3320.30 |    2.395 |    22.569 | E:	estfile.dat (30GB)
     5 |      8283185152 |      1011131 |      26.23 |    3357.20 |    2.375 |    21.607 | E:	estfile.dat (30GB)
     6 |      7820320768 |       954629 |      24.76 |    3169.60 |    2.508 |    21.745 | E:	estfile.dat (30GB)
     7 |      7896784896 |       963963 |      25.00 |    3200.59 |    2.479 |    21.981 | E:	estfile.dat (30GB)
-----------------------------------------------------------------------------------------------------
total:       64582090752 |      7883556 |     204.49 |   26175.34 |    2.428 |    22.258
Read IO
thread |       bytes     |     I/Os     |     MB/s   |  I/O per s |  AvgLat  | LatStdDev |  file
-----------------------------------------------------------------------------------------------------
     0 |      7940784128 |       969334 |      25.14 |    3218.43 |    2.471 |    22.884 | E:	estfile.dat (30GB)
     1 |      8152604672 |       995191 |      25.81 |    3304.28 |    2.401 |    22.211 | E:	estfile.dat (30GB)
     2 |      8116256768 |       990754 |      25.70 |    3289.55 |    2.408 |    22.080 | E:	estfile.dat (30GB)
     3 |      8180006912 |       998536 |      25.90 |    3315.38 |    2.394 |    22.936 | E:	estfile.dat (30GB)
     4 |      8192147456 |      1000018 |      25.94 |    3320.30 |    2.395 |    22.569 | E:	estfile.dat (30GB)
     5 |      8283185152 |      1011131 |      26.23 |    3357.20 |    2.375 |    21.607 | E:	estfile.dat (30GB)
     6 |      7820320768 |       954629 |      24.76 |    3169.60 |    2.508 |    21.745 | E:	estfile.dat (30GB)
     7 |      7896784896 |       963963 |      25.00 |    3200.59 |    2.479 |    21.981 | E:	estfile.dat (30GB)
-----------------------------------------------------------------------------------------------------
total:       64582090752 |      7883556 |     204.49 |   26175.34 |    2.428 |    22.258
Write IO
thread |       bytes     |     I/Os     |     MB/s   |  I/O per s |  AvgLat  | LatStdDev |  file
-----------------------------------------------------------------------------------------------------
     0 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
     1 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
     2 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
     3 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
     4 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
     5 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
     6 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
     7 |               0 |            0 |       0.00 |       0.00 |    0.000 |       N/A | E:	estfile.dat (30GB)
-----------------------------------------------------------------------------------------------------
total:                 0 |            0 |       0.00 |       0.00 |    0.000 |       N/A
  %-ile |  Read (ms) | Write (ms) | Total (ms)
----------------------------------------------
    min |      0.068 |        N/A |      0.068
   25th |      0.261 |        N/A |      0.261
   50th |      0.274 |        N/A |      0.274
   75th |      0.305 |        N/A |      0.305
   90th |      0.413 |        N/A |      0.413
   95th |      3.097 |        N/A |      3.097
   99th |     57.644 |        N/A |     57.644
3-nines |    198.563 |        N/A |    198.563
4-nines |    995.725 |        N/A |    995.725
5-nines |   1896.496 |        N/A |   1896.496
6-nines |   1954.282 |        N/A |   1954.282
7-nines |   1954.318 |        N/A |   1954.318
8-nines |   1954.318 |        N/A |   1954.318
9-nines |   1954.318 |        N/A |   1954.318
    max |   1954.318 |        N/A |   1954.318

The important part of this shows that at 204MB/s throughput and 26k IOPs, I had average 2ms latency.

thread |       bytes     |     I/Os     |     MB/s   |  I/O per s |  AvgLat  | LatStdDev
----------------------------------------------------------------------------------------
total:       64582090752 |      7883556 |     204.49 |   26175.34 |    2.428 |    22.258

Here is a view from my monitoring software, essentially validating the latency.

A good starting point for SQL workload testing would be something like:

diskspd –b8K –d30 –o4 –t8 –h –r –w25 –L –Z1G –c20G D:iotest.dat > DiskSpeedResults.txt

At this point, I just need to get the SSD installed on the host and test VMware Flash Read Cache.
To be continued…

Raid and IOPs Cheat Sheet – Excel

I made this 5 years ago in Excel and it was pretty popular. I think there are some errors in the math formulas that were pointed out for RAW calculations of backend IOPs. So use with caution, but mostly it should give a good enough idea on whatever it is you might be doing.

If you make modifications, please drop me a line in the comments or get in touch with me. I’ll update my post to include fixes.

RAID-and-IOPS Cheat Sheet Download

SQL Server: Cumulative IO By Database since Last Restart

Here’s a TSQL script to get cumulative IO by database since last restart.

WITH Agg_IO_Stats
AS
(
  SELECT
    DB_NAME(database_id) AS database_name,
    CAST(SUM(num_of_bytes_read + num_of_bytes_written) / 1048576 / 1024.
         AS DECIMAL(12, 2)) AS io_in_gb
  FROM sys.dm_io_virtual_file_stats(NULL, NULL) AS DM_IO_Stats
  GROUP BY database_id
),
Rank_IO_Stats
AS
(
  SELECT
    ROW_NUMBER() OVER(ORDER BY io_in_gb DESC) AS row_num,
    database_name,
    io_in_gb,
    CAST(io_in_gb / SUM(io_in_gb) OVER() * 100
         AS DECIMAL(5, 2)) AS pct
  FROM Agg_IO_Stats
)
SELECT R1.row_num, R1.database_name, R1.io_in_gb, R1.pct,
  SUM(R2.pct) AS run_pct
FROM Rank_IO_Stats AS R1
  JOIN Rank_IO_Stats AS R2
    ON R2.row_num <= R1.row_num
GROUP BY R1.row_num, R1.database_name, R1.io_in_gb, R1.pct
ORDER BY R1.row_num;

Source

How Do I Determine Application IOPS?

This is just a note for myself since I often find myself trying to analyze an application to performance tune a server/system.
Using Windows Performance Monitoring I use the following metrics when analyzing an application.

Metric	Example Data	Description
Logical Disk: Avg. Disk Bytes/Read	0.00	IO Size Read (Data block size)
Logical Disk: Avg. Disk Bytes/Transfer	8192.00	IO Size
Logical Disk: Avg. Disk Bytes/Write	8192.00	IO Size Read (Data block size)
Logical Disk: Disk Read Bytes/sec	0.00	Total Read data bytes per second
Logical Disk: Disk Write Bytes/sec	125000.00	Total Write data bytes per second
Logical Disk: Disk Transfers/sec	15.258	Total IOPS
Logical Disk: Disk Reads/sec	0.00	Read IOPS
Logical Disk: Disk Writes/sec	15.258	Write IOPS

To calculate Write IOPS for the application take “Disk Write Bytes/sec” and divide by “Avg. Disk Bytes/Write”.
125000.00 / 8192.00 = 15.258 IOPS
Notice that Disk Reads/sec and Disk Writes/sec corresponds to Read IOPS and Write IOPS already.