Search
StarWind is a hyperconverged (HCI) vendor with focus on Enterprise ROBO, SMB & Edge

Real-World Production Scenario: Squeezing all possible performance out of Microsoft SQL Server AGs (Availability Groups). Local NVMe Vs. RDMA-mapped (NVMe-oF) remote NVMe storage. Does the juice worth a squeeze?

  • December 23, 2021
  • 23 min read
StarWind DevOps Team Lead. Volodymyr possesses broad expertise in virtualization, storage, and networking, with exceptional experience in architecture planning, storage protocols, hardware sourcing, and research.
StarWind DevOps Team Lead. Volodymyr possesses broad expertise in virtualization, storage, and networking, with exceptional experience in architecture planning, storage protocols, hardware sourcing, and research.

Introduction

Sooner or later, every system administrator finds the need to have some data transferred over the network. And, of course, every system administrator usually knows how this goes, all pros and cons, which aren’t many, to be fair. Since the 2000s, one of the few popular and widely used solutions for such operations is iSCSI, a protocol for storage networking. Due to the latency, enterprises tend to apply iSCSI as secondary block storage. Essentially, it means that iSCSI has been used when the situation requires good performance, but it’s not vital. Common sense dictates that under such circumstances critical applications in dire need of the fastest performance should be run, for example, on a physically separate storage network. At least, that’s how it used to be.

The NVMe-oF protocol specification cannot offer much difference in terms of architecture or approach, but it does differ in the only way that matters: speed. As it has already been established at the example of Linux and StarWind solutions, the NVMe-oF initiators are capable of providing performance that is barely affected by latency. Strictly speaking, in the case with StarWind NVMe-oF Initiator for Windows, it would have hardly mattered for a lot of people whether the performance level could match local storage or not, since, in the absence of the Windows-native solution from Microsoft, the only alternative available is no NVMe-oF at all.

Purpose

StarWind NVMe-oF Initiator for Windows makes NVMe-oF available for SQL and M&E workloads on Windows and is compatible with all the major NVMe-oF storage vendors. Unfortunately, for enterprise-level IT infrastructures, all of these prospects are secondary in comparison with the performance. When the data transfer speed is critical to the results, every single microsecond can make all the difference, and ensuring the best performance and lowest latency for the users’ mission-critical applications is the number one priority.

That’s why we are going to use the high load Microsoft SQL Server (Always On Availability Groups) so that we could see for ourselves whether StarWind NVMe-oF Initiator for Windows is capable of handling the level of performance of local storage.

Benchmarking Methodology, Details & Results

For this benchmark, we will use two nodes, sw-mssql-01 and sw-mssql-02. These nodes will have two virtual disks respectively (this is where our test database goes):

Storage Spaces Simple from 2x Intel® Optane™ SSD DC P5800X Series (hereinafter referred to as local storage);

SPDK raid0 from 2x Intel® Optane™ SSD DC P5800X Series (hereinafter referred to as remote storage).

Testbed:

Testbed

Testbed architecture overview:

Microsoft SQL Server nodes:

Hardware:

Software:

Role Microsoft SQL Server node
OS
Windows Server 2019 Standart Edition (Version 1809)
MSSQL
Microsoft SQL Server Enterprise Evaluation (64-bit)
(Version 15.0.2000.5)
StarWind NVMeoF initiator
Version 1.9.0.471
StarWind Virtual SAN Version 8.0.14120.0

Client node:

Hardware:

sw-mssql-cli-01 Supermicro (SYS-2029UZ-TR4+)
CPU Intel(R) Xeon(R) Platinum 8268 CPU @ 2.90GHz
Sockets 2
Cores/Threads 48/96
RAM 96Gb
NICs 1x ConnectX®-5 EN 100GbE (MCX516A-CCAT)

Software:

Role Domain controller,  Client node
OS
Windows Server 2019 Standard Edition (Version 1809)
HammerDB
Version 4.2

Storage nodes:

Hardware:

sw-storage-{01..02} Supermicro (SYS-220U-TNR)
CPU Intel(R) Xeon(R) Platinum 8352Y @2.2GHz
Sockets 2
Cores/Threads 64/128
RAM 256Gb
Storage 2x Intel® Optane™ SSD DC P5800X Series (400GB)
NICs 2x ConnectX®-5 EN 100GbE (MCX516A-CDAT)

Software:

Role Storage node, NVMeoF target node
OS
CentOS Linux release 8.4.2105 (kernel 5.13.6-1.el8)
SPDK
SPDK v21.07 git sha1 adeb0496

The benchmark will be held using both DISKSPD and HammerDB utilities. It will be performed in separate stages. At first, we are going to see how different the performances of remote and local storage are using the DISKSPD utility. Furthermore, there comes the comparison of performances of both Microsoft SQL Server Standalone Instance and Microsoft SQL Server Availability Groups at the examples of local storage and remote storage using the HammerDB utility.

Preparation:

1.Add the nodes (sw-mssql-01 and sw-mssql-02) to the cluster using the Microsoft Failover Cluster:

Microsoft Failover Cluster

2. Create a fault tolerant virtual disk to act as a quorum using StarWind Virtual SAN:

StarWind Virtual SAN

Failover Cluster Manager

3. Create Storage Spaces Simple virtual disks on both nodes (sw-mssql-01 and sw-mssql-02):

Storage Spaces Simple

4. Connect remote disks from the nodes sw-storage-01 and sw-storage-02 to the nodes sw-mssql-01 and sw-mssql-02 respectively using StarWind NVMeoF Initiator;

5. Format disks to NTFS format with the allocation unit size 64K (R: and L: disks).

L: – local disk, Storage Spaces Simple form 2x Intel® Optane™ SSD DC P5800X Series;

R: – remote disk, SPDK raid0 from 2x Intel® Optane™ SSD DC P5800X Series.

Benchmarking Results:

DISKSPD:

This benchmark serves to define the performance of the local storage (Storage Spaces Simple) and the remote one (SPDK RAID0), connected through the NVMe-oF Initiator. With the consideration of the fact that Microsoft SQL Server operates at the file level, the disks in NTFS have the allocation unit size 64K. Needless to say, the benchmark patterns and block sizes below are most common for Microsoft SQL Server storage operations.

Master Database Files (MDF) workloads:

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Random Read 8k 4 4 452210 3533 0,026 346726 2709 0,046 77% 77% 177%
8k 4 8 453187 3541 0,062 431432 3371 0,074 95% 95% 119%
8k 4 16 447659 3497 0,134 465596 3637 0,137 104% 104% 102%
8k 8 4 774466 6051 0,032 634054 4954 0,050 82% 82% 156%
8k 8 8 775776 6061 0,072 749618 5856 0,085 97% 97% 118%
8k 8 16 770262 6018 0,156 797587 6231 0,160 104% 104% 103%
8k 16 4 1023902 7999 0,047 986691 7709 0,065 96% 96% 138%
8k 16 8 1021816 7983 0,109 1052851 8225 0,121 103% 103% 111%
8k 16 16 1020694 7974 0,235 1080298 8440 0,237 106% 106% 101%
8k 32 4 923119 7212 0,104 968126 7563 0,130 105% 105% 125%
8k 32 8 933105 7290 0,240 971248 7588 0,259 104% 104% 108%
8k 32 16 930719 7271 0,515 985062 7696 0,516 106% 106% 100%
Write 8k 4 4 426253 3330 0,034 326654 2552 0,049 77% 77% 144%
8k 4 8 425424 3324 0,066 404428 3160 0,078 95% 95% 118%
8k 4 16 427347 3339 0,140 437662 3419 0,146 102% 102% 104%
8k 8 4 742174 5798 0,041 589258 4604 0,054 79% 79% 132%
8k 8 8 741386 5792 0,075 692854 5413 0,091 93% 93% 121%
8k 8 16 743606 5809 0,161 728766 5693 0,171 98% 98% 106%
8k 16 4 1030897 8054 0,062 882748 6896 0,072 86% 86% 116%
8k 16 8 1029863 8046 0,109 1033064 8071 0,124 100% 100% 114%
8k 16 16 1028386 8034 0,233 1063625 8310 0,240 103% 103% 103%
8k 32 4 874428 6831 0,110 959987 7500 0,133 110% 110% 121%
8k 32 8 874637 6833 0,256 974352 7612 0,260 111% 111% 102%
8k 32 16 877076 6852 0,547 983368 7683 0,518 112% 112% 95%

 

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Random Read/Write
60/40
8k 4 4 438321 3424 0,027 337876 2640 0,047 77% 77% 174%
8k 4 8 436505 3410 0,064 417883 3265 0,076 96% 96% 119%
8k 4 16 436222 3408 0,137 452008 3531 0,141 104% 104% 103%
8k 8 4 756500 5910 0,035 615369 4808 0,052 81% 81% 149%
8k 8 8 755582 5903 0,074 726659 5677 0,087 96% 96% 118%
8k 8 16 757605 5919 0,158 767868 5999 0,164 101% 101% 104%
8k 16 4 1026260 8018 0,059 966484 7551 0,066 94% 94% 112%
8k 16 8 1024645 8005 0,117 1052538 8223 0,121 103% 103% 103%
8k 16 16 1024153 8001 0,234 1081764 8451 0,236 106% 106% 101%
8k 32 4 898509 7020 0,118 957946 7484 0,132 107% 107% 112%
8k 32 8 897383 7011 0,249 967748 7561 0,262 108% 108% 105%
8k 32 16 901274 7041 0,532 983779 7686 0,518 109% 109% 97%
Read/Write
40/60
8k 4 4 428689 3349 0,029 334287 2612 0,048 78% 78% 166%
8k 4 8 430305 3362 0,065 411856 3218 0,078 96% 96% 120%
8k 4 16 429201 3353 0,140 443415 3464 0,144 103% 103% 103%
8k 8 4 749105 5852 0,041 607243 4744 0,053 81% 81% 129%
8k 8 8 749480 5855 0,074 711163 5556 0,090 95% 95% 122%
8k 8 16 753685 5888 0,159 744860 5819 0,170 99% 99% 107%
8k 16 4 1025172 8009 0,062 941043 7352 0,068 92% 92% 110%
8k 16 8 1024748 8006 0,109 1030678 8052 0,124 101% 101% 114%
8k 16 16 1024585 8005 0,234 1054612 8239 0,242 103% 103% 103%
8k 32 4 894711 6990 0,107 957053 7477 0,133 107% 107% 124%
8k 32 8 893407 6980 0,255 968267 7565 0,262 108% 108% 103%
8k 32 16 894168 6986 0,550 982139 7673 0,517 110% 110% 94%

 

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Random Read 64k 2 2 114379 7149 0,035 77272 4829 0,051 68% 68% 146%
64k 2 4 169787 10612 0,047 104163 6510 0,076 61% 61% 162%
64k 2 8 207476 12967 0,077 114407 7150 0,127 55% 55% 165%
64k 4 2 175748 10984 0,045 123222 7701 0,065 70% 70% 144%
64k 4 4 211912 13245 0,075 179875 11242 0,088 85% 85% 117%
64k 4 8 220444 13778 0,145 200082 12505 0,154 91% 91% 106%
64k 8 2 212480 13280 0,075 190394 11900 0,084 90% 90% 112%
64k 8 4 220489 13781 0,145 218856 13679 0,146 99% 99% 101%
64k 8 8 223010 13938 0,287 222861 13929 0,287 100% 100% 100%
64k 16 2 220504 13782 0,145 218002 13625 0,146 99% 99% 101%
64k 16 4 223087 13943 0,287 222842 13928 0,287 100% 100% 100%
64k 16 8 224165 14010 0,571 224318 14020 0,570 100% 100% 100%
Write 64k 2 2 99589 6224 0,040 68432 4277 0,058 69% 69% 145%
64k 2 4 130258 8141 0,061 98202 6138 0,081 75% 75% 133%
64k 2 8 144870 9054 0,110 109409 6838 0,145 76% 76% 132%
64k 4 2 131697 8231 0,061 102934 6433 0,077 78% 78% 126%
64k 4 4 145058 9066 0,110 138960 8685 0,115 96% 96% 105%
64k 4 8 149092 9318 0,214 146892 9181 0,217 99% 99% 101%
64k 8 2 145152 9072 0,110 137721 8608 0,116 95% 95% 105%
64k 8 4 149135 9321 0,214 146985 9187 0,217 99% 99% 101%
64k 8 8 150643 9415 0,425 148954 9310 0,429 99% 99% 101%
64k 16 2 149052 9316 0,214 146489 9156 0,218 98% 98% 102%
64k 16 4 150704 9419 0,424 149021 9314 0,429 99% 99% 101%
64k 16 8 151326 9458 0,846 149985 9374 0,853 99% 99% 101%

 

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Random Read/Write
60/40
64k 2 2 105280 6580 0,038 66838 4177 0,059 63% 63% 155%
64k 2 4 146524 9158 0,054 99106 6194 0,080 68% 68% 148%
64k 2 8 167497 10469 0,095 106669 6667 0,135 64% 64% 142%
64k 4 2 149380 9336 0,053 102116 6382 0,078 68% 68% 147%
64k 4 4 168289 10518 0,095 151131 9446 0,105 90% 90% 111%
64k 4 8 173522 10845 0,184 166523 10408 0,191 96% 96% 104%
64k 8 2 168425 10527 0,095 154355 9647 0,103 92% 92% 108%
64k 8 4 173468 10842 0,184 171797 10737 0,186 99% 99% 101%
64k 8 8 175417 10964 0,365 174604 10913 0,366 100% 100% 100%
64k 16 2 173547 10847 0,184 171233 10702 0,186 99% 99% 101%
64k 16 4 175469 10967 0,364 174646 10915 0,366 100% 100% 101%
64k 16 8 176204 11013 0,726 175733 10983 0,728 100% 100% 100%
Read/Write
40/60
64k 2 2 102746 6422 0,039 65836 4115 0,060 64% 64% 154%
64k 2 4 139590 8724 0,057 98132 6133 0,081 70% 70% 142%
64k 2 8 156579 9786 0,102 108948 6809 0,137 70% 70% 134%
64k 4 2 141753 8860 0,056 99204 6200 0,080 70% 70% 143%
64k 4 4 157032 9814 0,102 144583 9036 0,110 92% 92% 108%
64k 4 8 161420 10089 0,198 159504 9969 0,200 99% 99% 101%
64k 8 2 157168 9823 0,102 146102 9131 0,109 93% 93% 107%
64k 8 4 161491 10093 0,198 160198 10012 0,199 99% 99% 101%
64k 8 8 163141 10196 0,392 162680 10167 0,393 100% 100% 100%
64k 16 2 161430 10089 0,198 159668 9979 0,200 99% 99% 101%
64k 16 4 163124 10195 0,392 162597 10162 0,393 100% 100% 100%
64k 16 8 163949 10247 0,780 163610 10226 0,782 100% 100% 100%

 

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Random Read 512k 1 1 12709 6355 0,078 11243 5622 0,088 88% 88% 113%
512k 1 2 21231 10616 0,094 13760 6880 0,145 65% 65% 154%
512k 2 1 25089 12544 0,079 12441 6221 0,160 50% 50% 203%
512k 2 2 28130 14065 0,142 19346 9673 0,206 69% 69% 145%
512k 4 1 28130 14065 0,141 21930 10965 0,182 78% 78% 129%
512k 4 2 28136 14068 0,284 28064 14032 0,285 100% 100% 100%
Write 512k 1 1 12125 6062 0,082 9797 4899 0,102 81% 81% 124%
512k 1 2 17713 8856 0,113 14235 7117 0,140 80% 80% 124%
512k 2 1 18375 9187 0,108 10813 5407 0,184 59% 59% 170%
512k 2 2 18928 9464 0,211 18579 9289 0,215 98% 98% 102%
512k 4 1 18919 9459 0,211 17495 8747 0,228 92% 92% 108%
512k 4 2 18964 9482 0,421 18707 9354 0,427 99% 99% 101%

 

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Random Read/Write
60/40
512k 1 1 12473 6236 0,080 10577 5289 0,094 85% 85% 118%
512k 1 2 19104 9552 0,104 13909 6955 0,143 73% 73% 138%
512k 2 1 20808 10404 0,096 11827 5913 0,168 57% 57% 175%
512k 2 2 22753 11376 0,175 19614 9807 0,203 86% 86% 116%
512k 4 1 22748 11374 0,175 19415 9708 0,205 85% 85% 117%
512k 4 2 22798 11399 0,350 22502 11251 0,355 99% 99% 101%
Read/Write
40/60
512k 1 1 12330 6165 0,081 10216 5108 0,097 83% 83% 120%
512k 1 2 18519 9259 0,108 14007 7003 0,142 76% 76% 131%
512k 2 1 19756 9878 0,101 11461 5730 0,174 58% 58% 172%
512k 2 2 21126 10563 0,189 19587 9794 0,204 93% 93% 108%
512k 4 1 21122 10561 0,189 18755 9377 0,213 89% 89% 113%
512k 4 2 21177 10589 0,377 20934 10467 0,382 99% 99% 101%

As you can see clearly for yourself, there is an obvious difference in performance (~30%-40%) in the cases with the small blocks and short I/O queues. Whenever these parameters are growing, the performance is growing respectively and can match that of the local disk.

Log Database Files (LDF) workloads:

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Sequential Read/Write
60/40
60k 1 4 81106 4752 0,049 76590 4488 0,052 94% 94% 106%
60k 1 8 81496 4775 0,086 77620 4548 0,090 95% 95% 105%
60k 1 16 81416 4771 0,184 77437 4537 0,193 95% 95% 105%
60k 2 4 134992 7910 0,059 102707 6018 0,078 76% 76% 132%
60k 2 8 136012 7969 0,103 110104 6451 0,131 81% 81% 127%
60k 2 16 136218 7982 0,220 111734 6547 0,268 82% 82% 122%
60k 4 4 184695 10822 0,086 169596 9937 0,094 92% 92% 109%
60k 4 8 185479 10868 0,172 178144 10438 0,179 96% 96% 104%
60k 4 16 185681 10880 0,344 168821 9892 0,367 91% 91% 107%
60k 8 4 185540 10872 0,172 184214 10794 0,173 99% 99% 101%
60k 8 8 185741 10883 0,344 184516 10811 0,347 99% 99% 101%
60k 8 16 185800 10887 0,689 184645 10819 0,693 99% 99% 101%
Read/Write
40/60
60k 1 4 81341 4766 0,049 76219 4466 0,052 94% 94% 106%
60k 1 8 81197 4758 0,086 76710 4495 0,091 94% 94% 106%
60k 1 16 81012 4747 0,185 76767 4498 0,195 95% 95% 105%
60k 2 4 135505 7940 0,059 99364 5822 0,080 73% 73% 136%
60k 2 8 135981 7968 0,103 111631 6541 0,134 82% 82% 130%
60k 2 16 135945 7966 0,220 111341 6524 0,269 82% 82% 122%
60k 4 4 172336 10098 0,093 163242 9565 0,098 95% 95% 105%
60k 4 8 173004 10137 0,185 168799 9891 0,189 98% 98% 102%
60k 4 16 173185 10148 0,369 168330 9863 0,380 97% 97% 103%
60k 8 4 173090 10142 0,185 171893 10072 0,186 99% 99% 101%
60k 8 8 173293 10154 0,369 172174 10088 0,371 99% 99% 101%
60k 8 16 173384 10159 0,738 172229 10092 0,743 99% 99% 101%

 

Test file size 400GB Local disk (L:) Remote disk (R:) Comparison
Test pattern Block size Threads I/O depth I/Ops MBps Latency (ms) I/Ops MBps Latency (ms) I/Ops MBps Latency
Sequential Read/Write
60/40
120K 1 4 58483 6853 0,068 47247 5537 0,084 81% 81% 124%
120K 1 8 58715 6881 0,119 47223 5534 0,148 80% 80% 124%
120K 1 16 58421 6846 0,257 47235 5535 0,317 81% 81% 123%
120K 2 4 92995 10898 0,086 57116 6693 0,136 61% 61% 158%
120K 2 8 93690 10979 0,171 68485 8026 0,219 73% 73% 128%
120K 2 16 93825 10995 0,341 68661 8046 0,437 73% 73% 128%
120K 4 4 93767 10988 0,170 91312 10701 0,175 97% 97% 103%
120K 4 8 93856 10999 0,341 93015 10900 0,344 99% 99% 101%
120K 4 16 93875 11001 0,682 93047 10904 0,688 99% 99% 101%
120K 8 4 93887 11002 0,341 93019 10901 0,344 99% 99% 101%
120K 8 8 93905 11004 0,681 93061 10906 0,687 99% 99% 101%
120K 8 16 93872 11001 1,363 92723 10866 1,380 99% 99% 101%
Read/Write
40/60
120K 1 4 58810 6892 0,068 47131 5523 0,084 80% 80% 124%
120K 1 8 58754 6885 0,119 47189 5530 0,148 80% 80% 124%
120K 1 16 58650 6873 0,256 47070 5516 0,318 80% 80% 124%
120K 2 4 87330 10234 0,091 64138 7516 0,120 73% 73% 132%
120K 2 8 87345 10236 0,183 68456 8022 0,219 78% 78% 120%
120K 2 16 87442 10247 0,366 68572 8036 0,437 78% 78% 119%
120K 4 4 87404 10243 0,183 85696 10042 0,186 98% 98% 102%
120K 4 8 87473 10251 0,366 86725 10163 0,369 99% 99% 101%
120K 4 16 87516 10256 0,731 86778 10169 0,737 99% 99% 101%
120K 8 4 87516 10256 0,365 86690 10159 0,369 99% 99% 101%
120K 8 8 87548 10260 0,731 86764 10168 0,737 99% 99% 101%
120K 8 16 87529 10257 1,462 86260 10109 1,484 99% 99% 102%

Microsoft SQL Server local & remote storage performance (HammerDB):

This benchmark is meant to compare the performance of the Microsoft SQL Server for the OLTP workload in a scenario wherein the database is allocated both on the local and remote storages. The benchmark will be held to measure the performance of both Standalone Instance and Availability Groups.

Basically, an OLTP (online transaction processing) workload is usually identified by a database receiving both requests for data and multiple changes to this data from a number of users over time where these modifications are called transactions.

Every single transaction has a defined beginning point. It modifies the data within the database and either commits the changes or takes them back to the starting point (more info here). Although designing and implementing a database benchmark is quite a task to take upon, TPROC-C is the OLTP workload implemented in HammerDB derived from the TPC-C specification with modification to make running HammerDB straightforward and cost-effective on any of the supported database environments (details up here).

Now, let’s get back to work, shall we?

Building the Schema:

  • Benchmark Options: TPROC-C;
  • RDBMS: Microsoft SQL Server;
  • Database Name: BenchmarkDB;
  • Number of Warehouses: 5000;
  • Number of Virtual Users: 100.

TPROC-C

The amount of warehouses directly affects the database size, which is why it was picked specifically to match the size of available storage resulting in the database with 670GB size. To make sure the results remain untainted, we did a backup of the database and were recovering it before every single test.

Availability Group parameters:

  • Name: SW-MSSQL-AG;
  • Replicas: SW-MSSQL-01 and SW-MSSQL-02;
  • Availability Mode: Synchronous Commit;
  • Failover Mode: Automatic;
  • Listener Name: SW-MSSQL-AG;
  • Data Synchronization: Automatic seeding

Availability Group

Failover Cluster Manager

HammerDB  benchmark parameters:

  • RDBMS: Microsoft SQL Server;
  • Database Name: BenchmarkDB;
  • Use All Warehouses: TRUE;
  • Timed Driver Script: TRUE;
  • Ramp Up Time (Minutes): 3;
  • Test Duration (Minues): 5;
  • Number of Virtual Users: {100,150,200,400,600,800}.

In order to show how the performance and load are going through changes depending on RAM, here is additional benchmarking with Microsoft SQL Server being given 256GB, 128GB, and 64GB RAM respectively:

Microsoft SQL Server – 256GB RAM:

Standalone Instance Availability Groups
Local Storage (L:) Remote Storage (R:) Local Storage (L:) Remote Storage (R:)
Virtual Users NOPM NOPM % Virtual Users NOPM NOPM %
100 1 929 194 1 567 104 81% 100 1 521 866 1 095 800 72%
150 1 897 995 1 683 390 89% 150 1 646 905 1 127 067 68%
200 2 092 169 1 888 431 90% 200 1 796 542 1 326 668 74%
400 2 156 676 2 090 032 97% 400 1 979 146 1 857 830 94%
600 2 134 175 2 222 587 104% 600 2 086 215 1 863 549 89%
800 2 371 548 2 244 145 95% 800 2 053 846 1 955 081 95%
Virtual Users TPM TPM % Virtual Users TPM TPM %
100 4 438 847 3 604 163 81% 100 3 502 152 2 522 072 72%
150 4 365 824 3 873 318 89% 150 3 786 312 2 557 345 68%
200 4 810 597 4 344 514 90% 200 4 132 772 3 051 800 74%
400 4 959 970 4 807 505 97% 400 4 551 920 4 272 246 94%
600 4 925 801 5 128 099 104% 600 4 816 668 4 292 334 89%
800 5 483 501 5 191 699 95% 800 4 745 047 4 517 652 95%

NOPM – new order per minute;

TPM – Transaction per minute.

Storage throughput usage:

Storage throughput usage

Storage throughput usage

Storage throughput usage

Microsoft SQL Server – 128GB RAM:

Standalone Instance   Availability Groups
Local Storage (L:) Remote Storage (R:) Local Storage (L:) Remote Storage (R:)
Virtual Users NOPM NOPM % Virtual Users NOPM NOPM %
100 1 926 874 1 491 056 77% 100 1 539 503 1 015 689 66%
150 1 975 635 1 713 846 87% 150 1 651 494 1 040 035 63%
200 2 038 390 1 923 053 94% 200 1 770 625 1 213 270 69%
400 2 187 896 2 121 844 97% 400 1 882 403 1 713 550 91%
600 2 216 416 2 138 098 96% 600 1 980 678 1 810 855 91%
800 2 227 702 2 166 103 97% 800 2 087 819 1 898 871 91%
Virtual Users TPM TPM % Virtual Users TPM TPM %
100 4 432 816 3 429 186 77% 100 3 539 969 2 337 354 66%
150 4 544 250 3 943 020 87% 150 3 799 009 2 392 843 63%
200 4 688 079 4 426 693 94% 200 4 072 705 2 789 238 68%
400 5 032 145 4 879 487 97% 400 4 330 568 3 940 653 91%
600 5 115 181 4 922 189 96% 600 4 567 002 4 168 913 91%
800 5 156 986 5 017 358 97% 800 4 827 349 4 371 248 91%

NOPM – new order per minute;

TPM – Transaction per minute.

Storage throughput usage:

Storage throughput usage

Storage throughput usage

Storage throughput usage

Microsoft SQL Server – 64GB RAM:

Standalone Instance   Availability Groups
Local Storage (L:) Remote Storage (R:) Local Storage (L:) Remote Storage (R:)
Virtual Users NOPM NOPM % Virtual Users NOPM NOPM %
100 1 847 859 1 432 296 78% 100 1 475 082 976 524 66%
150 1 957 018 1 690 486 86% 150 1 588 708 1 049 699 66%
200 2 056 806 1 938 364 94% 200 1 563 825 1 199 712 77%
400 2 145 978 2 051 504 96% 400 1 810 109 1 566 695 87%
600 2 249 779 2 093 142 93% 600 1 785 595 1 717 231 96%
800 2 274 036 2 116 673 93% 800 1 924 572 1 880 321 98%
Virtual Users TPM TPM % Virtual Users TPM TPM %
100 4 249 193 3 295 441 78% 100 3 392 623 2 245 828 66%
150 4 500 177 3 886 407 86% 150 3 653 900 2 415 594 66%
200 4 770 051 4 458 108 93% 200 3 597 298 2 759 022 77%
400 4 937 081 4 719 667 96% 400 4 174 774 3 608 951 86%
600 5 181 575 4 819 719 93% 600 4 115 204 3 951 591 96%
800 5 253 317 4 884 747 93% 800 4 441 703 4 337 072 98%

NOPM – new order per minute;

TPM – Transaction per minute.

Storage throughput usage:

Storage throughput usage

Storage throughput usage

Storage throughput usage

The benchmarking results shown are clear and definite: as you can observe, with a large number of Virtual Users (100,150,200), the performance difference between the local and remote storages could go up to ~30%. However, if we are to enlarge that number to 400,600,800, the performance difference becomes almost insignificant and does not exceed ~5%-10%.

Read throughput comparison (Microsoft SQL Server 64GB, 128GB, 256GB of RAM):

Moreover, the shortage of RAM for SQL server practically does not have any impact on write operations, but it does have a huge impact on read operations.

wp-image-18086

Conclusion

To sum up, the results of testing both with the means of DISKSPD and HammerDB have shown a clear and coherent picture. In particular, in the case with the latter, it becomes evident StarWind NVMe-oF Initiator for Windows handles large amounts of Virtual Users (400,600,800) quite well considering that the difference in performance doesn’t exceed 5-10%. However, when it comes to the loads with short queues, the picture is entirely different. With less than a numerous amount of Virtual Users (100,150,200), the performance difference reaches about 30%! The DISKSPD shows nearly the same numbers, with performance difference while handling small loads going up to 30-40%. Nevertheless, there’s nothing mysterious in such disproportional results since it has fairly nothing to do with the solution itself.

Due to the specifics of Windows storage stack architecture, the StarWind NVMe-oF driver is written as a Storport miniport because it is the only way Microsoft would support a block storage driver. Are there other options, you may ask? Of course, there are. However, as you can rightly guess, there are upside downs to that as well. For example, a monolithic SCSI port could have been a great alternative, although it lacks proper documentation. A class driver is quite efficient and useful, but it brings another headache as the system will neither handle SCSI reservation properly nor allow putting CSV (Cluster Shared Volume) on top of it. Essentially, StarWind NVMe-oF Initiator for Windows provides a fully compatible and Windows-certified NVMe-oF specification. Its performance as of now matches the performance of local storage and is only limited from the system side.

This material has been prepared in collaboration with Viktor Kushnir, Technical Writer with almost 3 years of experience at StarWind.

Found Volodymyr’s article helpful? Looking for a reliable, high-performance, and cost-effective shared storage solution for your production cluster?
Dmytro Malynka
Dmytro Malynka StarWind Virtual SAN Product Manager
We’ve got you covered! StarWind Virtual SAN (VSAN) is specifically designed to provide highly-available shared storage for Hyper-V, vSphere, and KVM clusters. With StarWind VSAN, simplicity is key: utilize the local disks of your hypervisor hosts and create shared HA storage for your VMs. Interested in learning more? Book a short StarWind VSAN demo now and see it in action!