In early May of 2013 starting with “Performance Comparison of Windows Server 2012 Essentials on the HP ProLiant N40L MicroServer and the HP ProLiant N54L G7 MicroServer” I posted a number of blogs where I compared the performance of a number of machines using the Intel NAS Performance Toolkit v1.7.1 (“NASPT”) Tests.
Description of NAS Performance Toolkit (“NASPT”) Tests
The NAS Performance Toolkit (“NASPT”) was developed and described by Tony Bock, Mason Cabot, Frank Hady, and Matthew Shopsin of the Storage Technologies Group, Intel Corporation, in the paper Measuring and Improving Single-User NAS Performance.
Portions of their abstract states:
NAS devices are increasingly entering the home and small business as centralized storage resources for large collections of documents, pictures, music and videos. Increasingly these devices are used for more than background tasks like backup. Newer interactive usages, like media access/creation, expose the performance of the NAS directly to the user. Unlike the enterprise NAS, the home and small business NAS will be judged primarily by single user performance as seen in user wait time.
We introduce a new tool, the NAS Performance Toolkit (NASPT), uniquely built to measure the single user NAS Performance seen by a user of a mainstream personal computer. NASPT includes a wide range of workloads identified by our analysis of media, productivity and bulk data operations likely to drive single user NAS performance.
We’ve made NASPT very easy to use and freely available. ….
A number of tests are provided with the NASPT. Below is a listing of those tests used in this report.
HD Video Playback: Traced from a commonly available video playback application, this trace represents about ten minutes of 720p high definition MPEG-2 video playback. A single 1.3GB file is accessed sequentially with 256kB user level reads. As is true in many of the workloads the NAS itself sees smaller reads since the SMB client and file system break these 256kB requests into smaller requests.
- HD Video Record: This trace represents recording roughly fifteen minutes of a broadcast 720p MPEG-2. A single 1.6GB file is written sequentially with 256kB access. The bit rate is somewhat lower than the playback test, they contain different video.
- HD Video Play & Record: This test was algorithmically constructed from the above video playback and record traces. To combine we introduced a 50ms offset into the record stream then merged the two streams. The 1GB file represents four minutes twenty seconds of application run time. Because the two streams have differing bit rates and because of variation in original trace periodicity, there is not a strict alternation of accesses. About 20% of the transactions are sequential.
- Two HD Video Playback Streams: Constructed from two copies of the above HD Video Playback test, this trace transfers 1.4GB of data representing two video streams played back for about six minutes. Again, sometimes one stream will issue two transactions in rapid succession so about 18% of the transactions are sequential.
- Four HD Video Playback Streams: This workload is constructed from four copies of the video playback test. The 1.3GB trace represents about three minutes forty-five seconds of video playback for each stream. About 11% of the accesses are sequential.
- Content Creation: This is a trace of commercially available video and photo editing software products executing a scripted set of operations to produce a video from a collection of different source materials. It contains a single very large file, apparently containing the video output, which is written in bits and pieces. About 11% of accesses within this file are sequential. There are many smaller files that are read and written more or less sequentially. Overall, about 40% of the accesses are issued sequentially. The test transfers 155MBs, 90% of transactions are writes. The median read size is 1300 bytes. The median write is 12kB. Transfers include a wide range of different sized accesses.
- Office Productivity: Scripted sequences of typical workday operations from a commonly available office productivity suite make up this trace. This test is the largest of the collection, transferring 2.8GB of data evenly divided between reads and writes. Eighty percent of these accesses are logically sequential, scattered across six hundred files ranging from 12 bytes in length to over 200MB. The median read size is 2.2kB whereas the median write size is 1.8kB.
- File Copy To NAS: This trace includes accesses executed when copying a 1.4GB file to a NAS. Data is written in 64kB sequential transactions.
- File Copy From NAS: Identical to File Copy To NAS, but in the opposite direction. All transactions are sequential 64kB reads.
- Directory Copy To NAS: This trace represents a bulk copy of a complex directory tree containing 2833 files, a transfer a large collection of files to the NAS. The directory used represented a typical installation of a commercially available office productivity suite. 247MBs is transferred with an average write size of 41.4kB. Only 52% of the writes are logically sequential as many files are small.
- Directory Copy From NAS: Identical to File Copy To NAS, but in the opposite direction creating many read accesses.
I recently decided to NASPT my Gen8 MicroServer Configuration described in Windows Server 2012R2 installed on my Gen8 MicroServer ODD SATA and the performance described in the HSS Forum thread Windows Server 2012R2 on RAID0 SSD on ODD SATA Port 5 of G8 MS.
Figure 1 -- NASPT Results for Gen8 MicroServer
For comparison, I ran NASPT for WHS-2011 running on my
Figure 2 -- NASPT Results for WHS-2011 running on Gigabyte X58A-UD3R
Additionally, I wanted to compare my recent NASPT Gen8 and X58 results to the May 2013 NASPT results running WHS-2011 on a N40L and N54L MicroServers.
Note: these NASPT Benchmarks are useful for Comparisons with My specific testing setup, network, and Client Workstation implementation and should not be considered absolute benchmarks for comparison against another person’s results or Comparisons made after a long separation of time, updates, and cumulative modifications. YMMV.
Figure 3 -- Comparison NASPT Results for N40L, N54L, X58, and Gen8
Graphing these results to compare in a bar chart.
Figure 4 -- Bar-Chart Comparison NASPT Results for N40L, N54L, X58, and Gen8
Figure 5 -- Comparison of NASPT Results for N40L, N54L, X58, and Gen8 -- Normalized to the N40L's Performance
In all categories my Gen8 MicroServer out performs my older generation N40L and N54L MicroServers and closely matches my X58A-UD3R performance while exceeding the X58A-UD3R’s performance in two categories.
Current As-Built & As-Tested: Machine
HP ProLiant Gen8 MicroServer Xeon E3-1265Lv2
16GB ECC RAM (Kingston KVR1333D3E9SK2/16)
OS: Microsoft Windows Server 2012R2 Standard
OS Drive C: 240GB Logical Drive 01 made of 1 x Corsair GT 240GB SSD in RAID0 on HP Dynamic Smart Array B120i RAID Controller (NOTE: this is attached to the ODD SATA Port 5 – SATA II or 3 Gbps)
Data Drive: 2TB Logical Drive 02 made of 4 x 1TB WD10JUCT 2.5” HDDs in RAID10 on HP Dynamic Smart Array B120i RAID Controller (NOTE: two drives are in Bays 1 & 2 – SATA III or 6 Gbps; and two drives are in Bays 3 & 4 – SATA II or 3 Gbps)
Please join us in the HSS Forum NAS Performance Toolkit Tests on Gen8 MicroServer with any question or comments or more information about my tests as well as in the HSS Gen8 Forums and HSS Forums.
Performance Measurement of the HP ProLiant N40L and N54L G7 MicroServer and Other NAS Devices http://homeservershow.com/performance-measurement-of-the-hp-proliant-n40l-and-n54l-g7-microserver-and-other-nas-devices.html
Check out CSKenney’s Gen8 links thread in the HomeServerShow forums: http://homeservershow.com/forums/index.php?/topic/5639-proliant-microserver-gen8-links/
Check out the Gen8 Blog postings: http://homeservershow.com/tag/gen-8
Check out the MicroServer Blog postings: http://homeservershow.com/tag/microserver