Matrix Storage Performance
Covering the native command queueing side of things is part of a separate article on all the major on-board SATA controllers doing the rounds on motherboards just now. Time constraints stop me from investigating it thoroughly here. Simple disk transfer performance has been investigated however.Since the controller is on the ICH6 itself, it doesn't have to ride the PCI bus or any other limited conduit, meaning any connected disk array should be allowed to stretch its legs. The pair of Maxtor disks shouldn't threaten the upper limits of the controller's performance in terms of sustained transfer, given their specification, so it'll be impossible to test those limits that way. If we secure a quartet of 74GB Raptors, supporting NCQ as well as providing blistering transfer speed and random access performance to shame all others, we'll revisit the sustained peak performance limits of the new disk controller.
However, the burst speeds should give some indication of the performance the controller is able to absorb.
So for now, we'll benchmark the Matrix Storage system in a couple of ways. Firstly using a RAID0 volume covering the entire available space, some 466GB over the pair of disks. That gives the baseline performance. Then a pair of volumes are created, having destroyed the original RAID0 volume, one RAID0 and the other RAID1, to investigate the performance hit when using Matrix Storage to create disparate volume types on the same array of disks.
Here's the baseline RAID0 fullsize HDTach 3.0.0 graph.
229.5MB/sec burst speed means the controller has plenty of headroom when moving data from the disks to the controller host itself. An average of 96MB/sec is acceptable for the disks being used (you'll see the single disk results soon). 5% CPU utilisation is excellent, showing the driver is doing all it can to utilise DMA transfers, with 16.4ms measured random access time none too shabby at all. The big MaxLine III disks appear excellent on the surface.
Given those figures, lets see what a RAID0 volume, when paired with a RAID1 volume on the same array, can do compared to that baseline.
The burst speed remains the same, maxing out the disks' burst ability. Sustained transfer rate drops into the 82MB/sec range, but random access time drops to 12.9ms and CPU utilisation is a low 4%. While the RAID1 array wasn't being made to work at the same time, it's clear that its presence has an effect on performance, even when it's idle.
Here's the RAID1 result.
54.6MB sustained with 130MB/sec burst speed confirms the RAID1 array's inherent performance differences, compared to RAID0. The random access time of 12.2ms shows you that RAID1 is good for read performance (reads can be done on either disk, the controller interleaving the commands to each disk in turn, maximising performance). CPU utilisation drops even further here, to a low 3%.
For reference, the single disk performance of the Maxtors were also recorded.
The burst speed narrowly beats that of a single 36.6GB Raptor, with the sustained read rate being very similar. The rotational speed of the Raptor, a heady 10,000rpm (compared to the Maxtor's 7,200rpm), means it outclasses the Maxtor in terms of random access performance.
The Raptor appears to run a little slower in terms of sustained read performance on the ICH6R, compared to other SATA controllers.