SPEED TESTS

FireWire Drives For Music - Part 1 - FireWire-less Macs

Published in SOS March 2002
Printer-friendly version

Technique : Computers

We recently tested the effect of processor upgrades on Mac music applications, but there are plenty of other factors that affect how well your audio software works. This two-part article shines the SOS spotlight on hard drive performance...

If you're a regular SOS reader, you may remember my article on processor upgrades for your older Macintosh from last December's issue (see www.soundonsound.com/sos/dec01/articles/macprocessor.asp). Upgrades like these provide a way of prolonging the life of your favourite beige-coloured box in the dog-eat-dog world of modern sequencers, but there's much more to running a sequencer efficiently than having a fast host processor. Processor upgrades can help you run more plug-ins, but when more tracks are required, it's your audio hard drive you need to look at.

FireWire is currently held to be the connectivity solution of choice for storage amongst computer-based musicians, but what if your Mac pre-dates FireWire and doesn't have it built in? The first of this two-part article on FireWire tests attempts to find out whether it's really worth adding FireWire to such older Macs and what kind of difference, if any, it can make to your sequencer's track count.

Most older Macs contain internal SCSI drives, and offer external SCSI connections. This used to be good news, but compared to today's IDE and, more recently, FireWire drives, upgrading SCSI drives can be expensive and problematic. Bewildering terminology abounds, such as wide SCSI, ultra-wide SCSI, ultra-fast-and-wide SCSI, eight-bit SCSI, 16-bit SCSI, SCSI I, II, and III, not to mention the range of possible connectors (25-pin, 50-pin and 68-pin, in case you were wondering). Not surprisingly, this is a science that not all musicians want to learn just to improve their track count, although take a look at Paul Wiffen's January Apple Notes column for a table that helps to make a lot of the definitions clearer (or surf to www.soundonsound.com/sos/jan02/images/applescsi.l.gif).

Test Spec     POWER COMPUTING POWERCENTER PRO 210 (CLONE) • CPU: 604e 210MHz. • Buss: 63MHz. • RAM: 192Mb. • Mac OS 9.1. APPLE G3 300 (BEIGE) • CPU: G3 300MHz. • Buss: 66MHz. • RAM: 256Mb. • Mac OS 9.1 APPLE G4 733 (SUPERDRIVE) • CPU: G4 733MHz. • Buss: 133MHz. • RAM: 256Mb. • Mac OS 9.1.

IDE is the cheapest form of hard drive available, and there are some very worthy examples out there with excellent performance (such as the IBM Deskstar 75GXP). Their low relative cost could be seen as an indicator of poor quality, but it has more to do with the fact that 99 percent of Wintel business and home machines are IDE equipped, which drives prices down. Installation of the drives themselves is not too tricky, but older Macs, such as the Power Computing Clone used in the tests this month, require the addition of a PCI-type IDE controller card. The Apple Beige G3 also used in this test has an IDE controller on the motherboard, so you would think that IDE might be the natural choice here, but ascertaining whether your particular Beige G3 supports a second IDE drive is a complex business (see the 'Shades Of Beige' box at the end of this article). And that's before you start to consider the installation, which is often fiddly and involves messing around with power lines inside your Mac.

At this point, owners of Macs with native FireWire ports are no doubt leaning back and looking smug, safe in the knowledge that they can add a 60Gb drive to their system without so much as opening the casing. For those of us with older machines that require an upgrade before they can use FireWire, fitting a suitable PCI interface card to take advantage of this technology may be a smart move, since it will make it much easier to transfer drives (and therefore your work) to any newer machines you buy in years to come. Although I relish a challenge as much as anyone, sometimes I just want to plug and play, and only FireWire hard drives offer high-level performance with just one connection.

Using a PCI-connected FireWire drive in your old machine would also allow you to easily interface your old machine with laptops, which can't take the controller cards for either SCSI or fast IDE drives. Furthermore, even after you've moved on to a shiny new Mac, your old machine can continue to be of use as a backup device, since two FireWire-equipped Macs can be connected with a FireWire cable, using 'target mode', where the desktop of one machine can be merged with the other for easy transfer of data.

In this article, I'll be looking at whether FireWire is a valid option for music use on older Macs by testing a LaCie 60Gb FireWire StudioDrive via two different PCI cards from Sonnet and Keyspan, on two different (very different, it turns out) age-challenged Macs. Next month, I'll look at testing Macs with built-in FireWire.

The LaCie drive is one of the latest generation of FireWire drives, which have overcome some of the problems of earlier models. A FireWire drive is essentially an ATA (aka IDE) drive in an enclosure with an IEEE1394 interface attached via a bridging chip. On earlier FireWire drives, this chip was a serious bottleneck to the performance of the drive, but this has now been resolved, and performance helped by the inclusion of a 2Mb cache. The FireWire interface has a potential transfer rate of 400Mbits per second, which translates to 50Mb per second (eight bits in a byte). With no other bottlenecks in the system, this ought to give very useable performance with a track count in excess of that required by most projects.

One potential bottleneck is the PCI FireWire adaptor card. Two types were available for test — the first, the FPCI3, is a three-socket FireWire-only affair from Keyspan, manufacturers of a range of PCI adaptor cards and other peripherals, who have put together an attractive package. This includes QuickTime Pro, which is worth $29.99 and is a very useful addition to any music computer. However, it should be noted that the bundled version is QuickTime Pro 4. As you may already have inadvertently found out (like I did), upgrading to QuickTime 5 results in the loss of your QuickTime 4 Pro key, and it's only recoverable by reinstalling QuickTime 4 and deleting QuickTime 5. If you want QuickTime 5 Pro, you need to purchase a new key, which is a bit much. None of this, of course, is Keyspan's fault.

The FPCI3 allows three FireWire devices to be connected externally as you might expect, or two externally and one internally — although an internal FireWire device is slightly pointless, since the only real advantage over IDE is 'hot-plugging'.

The G4/733 used for comparison purposes.

The second card is the Tango from Sonnet, whose processor upgrades cards fared well in the December 2001 article. The Tango is a combination card with two FireWire ports and two USB ports, which is an added advantage if you lack USB connectivity (true of all pre-Blue and White G3 Macs) and are short of PCI slots.

I used two computers for my test; a Beige G3 MiniTower (hereinafter referred to as 'the Beige') and a PowerComputing 604e-based clone (or from now on, 'the Clone', to save space). I also had access to an Apple 733MHz G4 (right) to make comparisons with more modern hardware.

In addition, I made use of some of the processor upgrade cards which I still had to hand from my December article, to see whether processor speed has any bearing on hard drive performance at the limit.

I searched the Internet for some time before commencing testing, trying to find a reliable benchmarking utility suitable for use with all types of drive, but to no avail (most programs have their detractors somewhere on the Internet). When the LaCie drive turned up with its own utility, TimeDrive, I thought my search was over. Fortunately, I had also designed a couple of tests of my own in Logic Audio, as these revealed TimeDrive to be a not entirely reliable benchmark! This conclusion was later borne out in a conversation with technical support staff at LaCie, who admitted that this particular program was not really up to scratch — although this was not the exact phrase they used...

Hard drives perform many operations during the delivery and storage of data, all of which can be measured. Some operations are mechanical, some are electronic and all have parameters that can be measured, which mean something to the people who design them but hardly anyone else. However, two parameters which do correlate to absolute performance as far as audio is concerned are seek time and sustained data transfer rate. The former is the average time it takes the head to find the data on the disk, and the latter is the speed with which the drive can deliver the data once the read head has found it. A typical song might contain references to many different audio files, located in different areas of the drive platter (testing the drive's seek time), while some of the files might be long takes with no break (testing the drive's data transfer rate).

The 'long file' audio test in Logic Audio's Arrange Window. The 'short file' test.

In order to differentiate between the two parameters, I devised two separate test files in Logic, the first based around continuous ('long') audio files, the other based on repeated short audio files. All the test audio was in the form of stereo, 44.1kHz, 16-bit files, due to the source material, and the audio hardware used was exclusively Mac AV in all cases.

The 'long files' audio test was constructed by converting the audio tracks from a sample CD to AIFF format using SoundApp (an extremely useful freeware batch-converting utility available at http://www-cs-students.stanford.edu/~franke/SoundApp/). I then burnt these files to a new CD, and copied the contents to each hard drive being tested. I loaded the AIFFs into Logic, with one audio file per audio track.

The 'short files' audio test was created by starting with a Logic song constructed with 24 tracks of one of the 'long' files mentioned above. Each one of these regions was chopped into 16th-note regions. Then, using Logic's 'Convert Regions into Individual Audio files' function (in the Audio menu), I converted each region into an actual audio file, so that each track consisted of 16 separate audio files per bar per track. This test attempts to recreate the kind of demand imposed on a hard drive by the playback of tracks made out of individual drum hits, for example.

Since these tracks were inconvenient to 'load' a track at a time during testing, several more tracks than the anticipated maximum track count were included with each test song; the test in each case consisted of running the song, stopping, unmuting the next track, and starting again. I repeated this process until Logic alerted me that the hard drive could no longer cope, then remuted the most recently muted track and, putting Logic into Cycle mode, made certain that the drive could indeed handle this many tracks on a long-term basis.

The TimeDrive test results in Table 1, below, are included merely to illustrate the benchmark problem. That is, the results compare reliably with the 'real world' tests in Table 2, but not in one or two cases. This benchmark appears to be accurate, but is not to be trusted as the sole indicator of performance. The results are given in Megabytes per second, and 'Error 36' is a TimeDrive error message whose meaning I have not been able to ascertain. The 'Read' and 'Write' measurements refer to readings taken with a single file size, while 'Variable Read' and 'Variable Write' refer to an average of readings taken with varying file sizes.

The Logic Audio track count tests are shown in Table 2 (on the next page). The results are given as the maximum possible number of tracks achieved by Logic in each case, apart from the Folder Copy results, which are given in minutes and seconds. The Internal drives carried System data as well as the test audio data, and were defragmented using Norton's Speed Disk prior to testing. In Logic's Audio Hardware and Drivers control panel the options pertinent to the track count were set as follows: Larger Process Buffer On, Larger Disk Buffer Off.

The Audio SCSI drive mentioned in the top line of Table 2 was a Quantum Fireball ST3.2 fitted as an audio-only (non-system) drive in this machine. 'a/m' indicates that a so-called 'Audio/MIDI sync error' message was flagged up by Logic. Finally, you'll see that there are some results at the end of the table for machines not described in this article (G4 867, iBook 500, iBook 600). These are machines used in the second part of this article, and the results are included here for the purposes of comparison next month.

The first point to note from the above results is that 'benchmarks' are not as reliable an indicator of performance as one might wish. For example, the highest-performing combination in the Beige running the Logic tests (LaCie/Tango/Sonnet G3/500), gives some of the poorest figures in the benchmark tests! The second point of note is that the Clone seems entirely unsuited to the addition of FireWire drives via the PCI buss when these are used for audio; the tests either gave audio/MIDI sync errors, or resulted in unreliable behaviour. There are a number of reasons why this might happen:

• Hardware conflicts between the PCI card and third-party (non-Apple) parts in the Clone. The Power Computing motherboard was originally an Apple design, but it has been 'chipped' to run a system buss speed of 60MHz. While the system itself is stable, this fact can lead to instability when other devices such as PCI cards are added in. The effect is similar to overclocking a processor; the more you diverge from the recommended speed, the more unstable they become.

• Other components in the system are simply not able to handle the data throughput that the software is demanding from the FireWire drive. For example, this computer has a 604e processor clocked at 210MHz. While the raw speed is more than two-thirds that of the Beige, the design of the processor and its caching system may not be up to the job in hand. The G3 chip was the first to include a 'backside' cache, an on-chip cache running at half the speed of the main processor, and capable of very fast data transfer, partly due to its proximity. The 604e relies on motherboard cache, whose transfer speed is slower. The error message is more likely to point to a conflict in the system, since slower architecture without conflict would simply result in reduced performance (in this case, lower track count).

Results for the Beige, conversely, are impressive, to say the least. Track count with long audio files is more than doubled, putting it in contention with the G4/733MHz Superdrive machine, and, surprisingly, the Beige gives better performance than the LaCie drive when connected to the native port of the Superdrive. With short files, the audio performance seems limited by processor speed, although this is not the only factor at play.

One particular combination seems to shine, that of the Sonnet Tango FireWire card with the Sonnet G3 500 ZIF processor upgrade. This should come as no surprise, bearing out the comments made above on hardware conflicts, and the fact that both these components come from the same company. For the same reason, it is easier to understand the poor performance of the Keyspan/Sonnet processor upgrade combination in the short audio files test.

One further point of interest is that even though the G4/733 internal drive scores over the LaCie in the long audio files test, the FireWire drive has a better seek time and wins the short files test.

As with processor upgrades, the age of your machine seems to be a governing factor in the decision to spend money in order to improve performance. Despite my unsuccessful experiences with the Clone, it is possible that PCI FireWire cards may give useable results for audio in some pre-G3 machines, and I would be interested to hear from anyone with machine-specific experience.

Upgrading your processor as well as adding hard drive options makes a lot of sense. The Keyspan card gives useable results and is cheaper than the Sonnet, but be aware that conflicts can arise with certain makes of processor upgrades.

Not surprisingly, the Sonnet combination of processor upgrade and FireWire PCI card gave the best results, and comes highly recommended. Good news at last — modern-day performance is possible for a fraction of the cost of a new machine.

Shades of Beige — Pre-Blue and White G3 Types     Sold from November 1997 to January 1999, the beige Apple G3 came in three basic types — the desktop, the mini-tower and the server. The Desktop (DT) model relied on the ultra compact 'Gossamer' motherboard to mount the hard drive on the floor of the case, leaving space for an optional internal Zip drive plus one further drive. Height restrictions in the case design restricted the maximum RAM to 192Mb, although this situation was resolved with the release of low-profile RAM DIMMs. The Mini-Tower (MT) version had a hinged case similar to the 8600/9600 design, but shorter. This design provided plenty of space and access for easy upgrading, additional drives, and so on. The Mini-Tower offered Video input/output in addition to the Audio input/output of the desktop, both being provided by 'personality' cards mounted on the motherboard. The Server option, while similar in case design to the Mini-Tower, lacks the personality card, and was originally sold with extra 'server' software — although no hardware marks out these machines as servers! The hard drive options are confusing. The Gossamer motherboard features an onboard IDE controller, and although the table on the right lists the original specification of each model, it should be noted that SCSI was an option (PCI-controlled) on IDE-equipped machines, and vice versa. In addition, machines with a 'Revision 2' motherboard (most easily distinguishable by the fact that the System CD supplied was blue, not white, and the Revision 2 board had ATI Rage Pro graphics where the Revision 1 had ATI Rage IIc) had a socket for controlling internal SCSI drives on the motherboard, and therefore did not require the PCI controller. Furthermore, only the Revision 2 motherboard could support an IDE drive as a second drive (or slave). So both revisions could have either IDE or SCSI as standard, or various combinations — depending on motherboard revision — as an option! These were the last machines offered by Apple with SCSI system drives (the Blue and White G3 was IDE only), and the first offered with IDE, so it is not surprising that hard drive type is a grey (or beige!) area. Machine Hard Drive Type Hard Drive Size Backside cache 'Personality' card G3/233 DT E-IDE 4Gb 512K Whisper G3/233 MT• E-IDE 4Gb 512K Whisper G3 Server 233 MT Ultra Wide SCSI 4Gb 512K n/a G3/266 DT E-IDE 4Gb 512K Whisper G3/266 MT E-IDE 6Gb 512K Wings G3 Server 266 MT Ultra Wide SCSI 4Gb 512K n/a G3/300 DT E-IDE 6Gb 1Mb Whisper G3/300 MT Ultra Wide SCSI 4Gb/8Gb 1Mb Wings G3 Server 300 MT Ultra Wide SCSI 4Gb 1Mb n/a G3/333 MT Ultra Wide SCSI 9Gb 1Mb Wings G3 Server 333 MT Ultra Wide SCSI 9Gb x 2 1Mb n/a DT = desktop. MT = minitower. • special order only. The 'Wings' personality card is Video and Audio input/output. The 'Whisper' personality card is Audio input/output only. Both Wings and Whisper support optional DVD/FireWire.

Pricing & Contacts     KEYSPAN • Keyspan FPCI3 FireWire PCI card: £69.33 AM Micros +44 (0)1392 426473. www.keyspan.com LACIE • LaCie 60Gb FireWire StudioDrive: £210.33 LaCie +44 (0)20 7872 8000. www.lacie.com SONNET • Sonnet Tango USB/FireWire combo PCI card: £93.99 Computers Unlimited +44 (0)20 8200 8282. www.sonnettech.com All prices include VAT.