메뉴 건너뛰기

2019.02.26 01:39

FIO Benchmark

조회 수 30 추천 수 0 댓글 0


Prev이전 문서

Next다음 문서

크게 작게 위로 아래로 댓글로 가기 인쇄


Prev이전 문서

Next다음 문서

크게 작게 위로 아래로 댓글로 가기 인쇄
Fio - Flexible I/O Tester Synthetic Benchmark

Fio is short for Flexible IO, a versatile IO workload generator. Back in 2005, Jens Axboe, the backbone behind and author of the IO stack in the Linux kernel, was weary of constantly writing one-off test programs to benchmark or verify changes to the Linux IO subsystem. As a result, fio was born to make the job a lot easier. It was flexible enough to allow detailed workload setups, and it contained the necessary reporting to make sense of the data at completion. Jens continued working on fio when he joined Oracle and, later, Fusion-io. Today, the fio community of users is active and engaged in development, and consequently they continually develop fio and implement new features. Greater than 100 people have contributed to fio, and many have done so several times. Thanks to this commitment, a new version of fio is released roughly every 4-6 weeks, and fio is widely used as an industry standard benchmark, stress testing tool, and for IO verification purposes.

Main Feature Set and Capabilities

The two primary features in a good benchmark are being able to run the workload that you want and getting the desired units of output. Being flexible was (and continues to be) the main focus of fio. It supports workload options not found in other benchmarks along with rigorously detailed IO statistics. Any sort of random and sequential IO mix, or read/write mix, is easy to define. The internal design of fio is flexible as well. Defining a workload is completely separate from the IO engine (a term fio uses to signify how IO is delivered to the kernel). For instance, if you want to run a workload with native async on Linux and then compare the same workload on Windows, just change the single IO engine line to the native version on Windows. Or, if you want more detailed latency percentile statistics at the tail of the distribution, that's easy too. Just specify the exact percentiles in which you are interested and fio will track that for you.

Though initially developed on Linux, the areas in fio that aren't tied to OS-specific features work on any platform from Windows to HP-UX to Android. There are also native IO engines on Linux, Windows, Solaris, etc. This is a key feature for mixed environments where you want to be able to run (as close to) identical workloads on different operating environments. Additionally, while fio is normally run directly on the target machine, it also supports network connections. You can run a server backend on the target machine(s) and run the data collection frontend on a client machine. This makes it easier to manage, especially if fio is often used on multiple machines.

Fio is predominantly a text-based CLI application, though initial support exists for a cross-platform gtk-based GUI frontend (gfio). At the time of this publication, gfio is included with the latest fio release, v2.1, and it should be fairly stable. It is able to serve as a GUI frontend for any workload that the CLI client supports. It also provides basic support for GUI editing job options and workloads, though full support is still in development as gfio is a work in progress.

Fio also supports three different types of output formats. The “classic” output is the default which dumps workload statistics at the end of the workload. There's also support for a CSV format, though that's slowly diminishing in favor of a JSON-based output format. The latter is far more flexible and has the advantage of being simple to parse for people and computers.

Fio has been moving forward rapidly thanks to the synergy of its vibrant user and development community. It's easier to run than competing projects, flexible enough to accomplish users' tasks, and it has a low enough overhead to drive any kind of storage system at full speed. Couple that with its richer options and reporting than anything else out there, and fio is an excellent tool.

Why StorageReview.com uses FIO

As we've begun increasing both the quantity and depth of our enterprise storage reviews, we've been in need of a better benchmarking tool to accurately measure the performance of different storage products across multiple operating systems. Traditional workload generator software such as Iometer or Vdbench offer limited compatibility with their non-native OS, or they've had difficulty scaling loads for high-performance storage devices such as PCIe application accelerators or high-performance networked storage. Since our implementation of FIO began, it has solved many of these problems and has even gone a step further by supporting scripting for lengthy testing periods. This overall functionality makes it our synthetic benchmark of choice for all enterprise reviews.

For our reviews, we use FIO to measure the performance of a storage device over a given period of time. For most products, that includes 6 hours of preconditioning before we drop into our main tests. For larger PCIe storage devices that may not enter steady-state for many hours into testing, we precondition for twice as long at 12 hours before our main tests commence. Our synthetic enterprise storage benchmark process begins with an analysis of the way the drive performs during a thorough preconditioning phase. Each of the comparable products are first secure erased using the vendor's tools and then preconditioned into steady-state with the same workload the device will be tested with under a heavy load of 16 threads with an outstanding queue of 16 per thread. Rounding out the process, we then test in set intervals in multiple thread/queue depth profiles to show performance under light and heavy usage.

Preconditioning and Primary Steady-State Tests:

  • Throughput (Read+Write IOPS Aggregate)
  • Average Latency (Read+Write Latency Averaged Together)
  • Max Latency (Peak Read or Write Latency)
  • Latency Standard Deviation (Read+Write Standard Deviation Averaged Together)

Our Enterprise Synthetic Workload Analysis includes different profiles to reflect some real-world tasks. These profiles have been developed to make it easier to compare to our past benchmarks as well as widely-published values such as max 4k read and write speed and 8k 70/30, which is commonly used for enterprise hardware.

  • 4k
    • 100% Read or 100% Write
    • 100% 4k
    • fio --filename=/dev/sdx --direct=1 --rw=randrw --refill_buffers --norandommap --randrepeat=0 --ioengine=libaio --bs=4k --rwmixread=100 --iodepth=16 --numjobs=16 --runtime=60 --group_reporting --name=4ktest
  • 8k 70/30
    • 70% Read, 30% Write
    • 100% 8k
    • fio --filename=/dev/sdx --direct=1 --rw=randrw --refill_buffers --norandommap --randrepeat=0 --ioengine=libaio --bs=8k --rwmixread=70 --iodepth=16 --numjobs=16 --runtime=60 --group_reporting --name=8k7030test

Currently we utilize FIO version 2.0.7 (x64) in CentOS 6.3 for our Linux performance tests and FIO version (x64) in Windows Server 2008 R2 SP1 for our Windows performance.

StorageReview Enterprise Testing Platforms

Storage solutions are tested with the FIO synthetic benchmark in the StorageReview Enterprise Test Lab utilizing stand-alone servers. We utilize the HP ProLiant DL380/DL360 Gen9 to show realistic performance, changing only the storage adapter or network interface to connect our DL380/DL360 to different storage products. The  DL380/DL360 have proven itself to offer great compatibility with third-party devices, making it an excellent go-to platform for this diverse testing environment. 

(Gen1) Lenovo ThinkServer RD240

  • 2 x Intel Xeon X5650 (2.66GHz, 12MB Cache)
  • Intel 5500+ ICH10R Chipset
  • Memory - 8GB (2 x 4GB) 1333Mhz DDR3 Registered RDIMMs
  • Windows Server 2008 Standard Edition R2 SP1 64-Bit and CentOS 6.2 64-Bit
  • LSI 9211 SAS/SATA 6.0Gb/s HBA

(Gen2) Lenovo ThinkServer RD630

  • 2 x Intel Xeon E5-2620 (2.0GHz, 15MB Cache, 6-cores)
  • Intel C602 Chipset
  • Memory - 16GB (2 x 8GB) 1333Mhz DDR3 Registered RDIMMs
  • Windows Server 2008 R2 SP1 64-bit, Windows Server 2012 Standard, CentOS 6.3 64-Bit
  • LSI 9207-8i SAS/SATA 6.0Gb/s HBA (For benchmarking SSDs or HDDs)

(Gen3) HP ProLiant DL380/DL360 Gen9

List of Articles
번호 분류 제목 글쓴이 날짜 조회 수
131 Linux CentOS 7 network prefix 맨하탄노숙자 2019.03.27 26
» Linux FIO Benchmark 맨하탄노숙자 2019.02.26 30
129 Linux 리눅스에서 하드디스크 이미지 저장 맨하탄노숙자 2018.12.29 37
128 Windows Ghost Solution Suite - Ghost 11.5.1 맨하탄노숙자 2018.10.03 508
127 Linux nmtui 맨하탄노숙자 2017.11.21 119
126 Linux xfs repair 맨하탄노숙자 2017.10.14 151
125 Linux RHEL7(CentOS7) 의 grub2 사용 해 보기 맨하탄노숙자 2017.07.26 234
124 Linux VMware ESXi SSD on RAID showing as non-ssd 맨하탄노숙자 2017.05.09 131
123 Windows Windows XP 의 윈도우에서 Visual Studio .net (2003) 설치시 옵션 맨하탄노숙자 2017.01.02 151
122 Windows 윈 10 인증 크랙 secret 맨하탄노숙자 2016.05.01 8
121 Windows 윈도우10 정품인증을 확실하게 받을 수 있는 방법입니다 맨하탄노숙자 2016.05.01 3683
120 Linux CentOS 7 CPU 퍼포먼스가 100% 안나올때 맨하탄노숙자 2016.03.25 541
119 Linux ## 아파치 설치용으로 CSR을 생성하고, 발급받은 VeriSign(베리사인) SSL 인증서를 Tomcat 8 에 적용 맨하탄노숙자 2016.03.22 1551
118 Linux mysql old_password 설정 맨하탄노숙자 2016.03.22 304
117 Windows 윈도우7 원격데스크톱 2인 이상 동시 접속 (다중 사용자 모드) 설정 방법 file 맨하탄노숙자 2016.03.21 772
116 Linux rsync 맨하탄노숙자 2016.03.21 210
115 Linux Rsync 특정 기간(일정 기간) 이상 지난 컨텐츠 복사 (730일, 365일, 1주일) 맨하탄노숙자 2016.03.10 264
114 Linux iptables 용어 맨하탄노숙자 2016.02.18 8134
113 Linux iptables 용어 맨하탄노숙자 2016.02.18 8291
112 Linux Linux iptables 방화벽 설정 맨하탄노숙자 2016.02.18 255
Board Pagination Prev 1 2 3 4 5 6 7 Next
/ 7