PERC6 with MD1000 and MD1120 Performance Analysis ReportThis is a featured page

The Dell™ System Performance Anaylsis team has taken another step in their PERC 6 performance analysis work and has included the new MD1120 in a report that covers a very large amount of data. The full paper is posted here in PDF form; this page has the summary and highlights. Please feel free to post questions about this work using the discussion threads at the bottom.

Dell PowerEdge Expandable RAID Controller 6 on the Dell PowerVault MD1000 and MD1120 Performance Analysis Report

This paper provides an analysis of the performance of the Dell PowerEdge™ Expandable RAID Controller 6 (PERC 6). The measurement of RAID controller performance is often referred to being an “art” as much as a “science.” Many factors come into play, including the controller itself, the number and type of drives that are attached, the OS being used, the I/O subsystem of the server, and the application in question.

Our performance analysis uses I/O file size, workload type, number of disks, cache enabling/disabling and RAID level as the input test parameters. The reader of this report is expected to be technical, wanting to collect more information and/or get an idea about the performance characteristics of this product.

Please see the PERC 6 user's guide for more product details: http://support.dell.com/support/edocs/storage/RAID/PERC6/en/PDF/en_ug.pdf. NOTE: The PDF download file is 32 MB.

Major Advantages

Using 3Gb/s SAS technology, PERC 6 was observed in testing to be able to transfer up to 1484 MB—the theoretical upper limit by RAID On Chip (ROC) is better than 1.5 GB—using sequential reads and employing both of the external channels. Streaming video and data backup applications are capable of running extremely fast with up to 1 MB stripe configurations. This measurement is compared to the previous generation PERC 5 with a maximum available 128 KB stripe size.

The PERC 6 supports all key RAID levels, including RAID 6 and 60. RAID 6 tolerates two simultaneous failures, and thus offers the ability to manage unrecoverable media errors during rebuilds. This ability is useful, especially when utilizing many high capacity, half-duplex, lower-cost Serial ATA (SATA) disks.

The battery-backed cache offers a large, addressable memory space that can increase throughput in several applications and preserve cache contents up to 24 hours using a 7 WH Lithium Ion rechargeable battery.


Test Environment

In the tests Dell Labs used the IOMETER benchmark tool to stress the disk sub-system. With IOMETER the tests generated 12 different types of workloads to represent different application patterns commonly used by Dell customers. Different application platforms pass data and commands
to the RAID controller in different block sizes. These blocks are called I/O files. To simulate different platforms, the tests used the payload sizes listed in the graphs presented later to represent a variety of applications using I/O sizes ranging from small to large. The payload sizes were split for graphing to optimize scale of the graph between small patterns with large I/Os per second (IOPS) scores, and large patterns with high MB/s but low IOPS scores.

Test Results

The following data is based on extensive PERC 6 IOMETER tests in Dell Labs. The graphs that follow are charted from data in Appendix A of the full paper.

The PERC 6 default operational mode is Write Back No Read Ahead mode because of lower latency times on write activity when in Write Back mode compared to Write Through mode, which will wait on the write transaction until the write is complete. For most customers the default settings are preferred, but for some customers the default adapter settings may not always offer best performance so Dell provides a choice.

The PERC 6 offers RAID levels with the choice of stripe sizes 8 KB, 16 KB, 32 KB, 64 KB, 128 KB, 256 KB, 512 KB, and 1 MB with the default being 64 KB. We chose 512 KB and tested RAID levels 0, 1, 5, 6, and 10. Higher stripe sizes create reduced physical I/O operations and enhance PCI-Express bus utilization. Stripe size 512 KB was selected because it is commonly used in the Linux® OS and was determined to have the best combination of benefit for large I/O and least detriment for small I/O on this controller across multiple operating systems. In the following tests, the number of drives used was determined by the maximum capacity of the storage enclosure under test. The virtual disk count in all tests was three or more on the Dell PowerVault™ MD1000 and at least four virtual disks on the PowerVault MD1120.

PowerVault MD1000 Results

The modular design of the Dell PowerVault MD1000 is engineered for easy expansion, providing additional room to store data from the server. Up to six PowerVault MD1000 15 mixed-drive expansion enclosures can be daisy-chained together on both channels, providing up to 90 mixed disk drives options. To keep management simple, the PowerVault MD1000 connects to and can be controlled by a single PERC RAID controller-equipped PowerEdge server. The following results reflect a single MD1000 with 15 hard disk drives. The queue depths increase from 2 to 64 outstanding I/Os.

The results for RAID 5 MB/s and IOPS are shown here. The results for RAID 0, 6, and10 are in the PDF of the full report.

PowerVault MD1000 RAID 5 MB/s

MD1120 RAID 5 MBps Results

PowerVault MD1000 RAID 5 IOPS

MD1120 RAID5 IOPs Performance Results

PowerVault MD1120 Results

The Dell PowerVault MD1120 is Dell’s newest storage enclosure. Up to six PowerVault MD1120 24 mixed-drive enclosures can be daisy-chained together on both channels, providing up to 144 mixed disk drives options. To keep management simple, the PowerVault MD1120 connects to and can be controlled by a single PERC RAID–controller equipped Dell PowerEdge server. The following results reflect a single MD1120 with 24 SAS 2.5” Seagate Savvio 15K.1 drives. The queue depths increase from 2 to 64 outstanding I/Os.

The results for RAID 5 MB/s and IOPS are shown here. The results for RAID 0, 6, and10 are in the PDF of the full report.


PowerVault MD1120 RAID 5 MB/s

MD1120 RAID5 MBps Performance Results

PowerVault MD1120 RAID 5 IOPS

MD1120 RAID5 IOPs Performance Results

Please refer to the complete version of theDell PowerEdge Expandable RAID Controller 6 on the Dell PowerVault MD1000 and MD1120 Performance Analysis Report for more results and testing details.

Post questions or comments below about the report.


No user avatar
TDA-Terry
Latest page update: made by TDA-Terry , Jan 15 2009, 8:17 PM EST (about this update About This Update TDA-Terry Edited by TDA-Terry

8 words added

view changes

- complete history)
Keyword tags: DAS MD1120 PERC PERC6 storage
More Info: links to this page
Started By Thread Subject Replies Last Post
DrInequality What a crock 0 Aug 17 2010, 10:12 PM EDT by DrInequality
Thread started: Aug 17 2010, 10:12 PM EDT  Watch
Approaching 2000 random IOps with small queue depths and rotating hard disks? I don't think so.

Individually the drives are 285 IOps. With small queue depths you can't do much better than that.

I call ********
Do you find this valuable?    
Keyword tags: DAS MD1120 PERC PERC6 storage
melzerpinto MD1000 0 Nov 4 2009, 4:15 PM EST by melzerpinto
Thread started: Nov 4 2009, 4:15 PM EST  Watch
Hi,
Has anyone tested the MD1000 throughput with a Poweredge R610? I've seen the test results conducted by Dell using iometer but I'm not able to get anything greater than 480 MB/s. The MD1000 has 10 drives configured in RAID10 and the OS is SLES 10 SP2. I'm using a single controller.I used Iometer like in the Dell tests.I'm using Perc 6/E and the virtual disk has a stripe size of 64k
When I use 2 controllers(same server) and stripe an LVM, I get about 630 MB/s which I still think is pretty low. The MD1000 is running in unified mode. If anyone has conducted tests, can they share their results?

Thanks
Melzer
Do you find this valuable?    
Keyword tags: DAS MD1120 PERC PERC6 storage
Joe_Liston Question on Raid10 1 Jul 31 2009, 4:24 PM EDT by Dan_Hambrick
Thread started: Jul 31 2009, 11:18 AM EDT  Watch
In this report you state for RAID 10 Analysis :"For example, for reads we will utilize only half the disks in the array since the other half are just mirrored copies." But in the overall analysis you list Raid 10 as on of the BEST for Random reads. I assume this is since you can read from both disks in a mirrored pair. Is this assumption correct? For Sequential reads does it keep the reads local to a single disk in a mirror?
Do you find this valuable?    
Keyword tags: None
Show Last Reply
Showing 3 of 3 threads for this page