starman

Burst Rate for U320 drives

Recommended Posts

I have built in U320 controllers in my gigabyte motherboard.

Using the HD Tech benchmark, I can only get up to 113.5MB/sec Burst Rate from my U320 SCA2 Cheetah 36G HD. I am using Cremax Icy Docks.

Is there a flaw in my desire to see 320MB/sec Burst Rate?

Share this post


Link to post
Share on other sites

If all of your other PCI slots are 32bit/33MHz, it is highly unlikely that the integrated components are anything else.

If that is the case, this would explain the low burst rate. The PCI bus is saturating.

Share this post


Link to post
Share on other sites

What's your board? There were two versions of this Gigabyte board (8KNXP Ultra or somesuch?) as far as I remember - the first one had SCSI connected to PCI32/33, and the later one ran it off PCI64/66. Don't remember all the details, Google should help..

Share this post


Link to post
Share on other sites
Why would you care about STR that high anyway?

Something I am rather fond of quoting from this SR page is

What ramifications does locality have on drive performance? It means that short-seek performance and especially buffer size in conjunction with caching strategy have a much larger effect on net hard drive performance than most folks realize. In fact, a drive's buffer and its accompanying read-ahead and write-back caching strategies exert more influence on net single-user performance than spindle speed, seek time, or transfer rates! A corollary accompanies this admission: If buffers and caching strategies effect significant performance increases, then buffer hits often occur. And if buffer hits occur often, buffer-to-host transfer rates, otherwise known as burst transfer rates, are significant.

Share this post


Link to post
Share on other sites

Your 8KNXP Ultra (even the 2.0) is only PCI 32/33.

64-bit PCI physically has a longer slot than 32-bit PCI so it's very easy to spot the difference. Virtually all desktop-market boards only have 32-bit PCI at this time. There's no on-board SCSI, so the "on-board SCSI connection to the PCI bus" is completely irrelevant. :)

http://www.giga-byte.com/MotherBoard/Produ...(Rev%202.0).htm

Share this post


Link to post
Share on other sites
In fact, a drive's buffer and its accompanying read-ahead and write-back caching strategies exert more influence on net single-user performance than spindle speed, seek time, or transfer rates!

Funny you mention that. I went from a dualie 2Ghz Athlon to a Opteron 148 single proc. My dualie was soooo much faster. I guess I've lost the single user aspect of myself.

Share this post


Link to post
Share on other sites

I spoke with Gigabyte customer service and they claim that this is the 1st time that they have heard of this low burst rate problem (ha!).

Customer service suggested that HD Tach was giving errant burst rate readings because it was taking them from the 33Mhz bus.

Any comments?

Share this post


Link to post
Share on other sites

Well the 32 bit PCI (or 33MHz PCI bus) of SCSI can only handle about 160 MB/s. But if yours was the 64 bit (66MHz PCI bus) of SCSI, then it should be bursting at 320 MB/s. Please remember though, just because the SCSI burst (assumed as being the 64 bit version) MAY support 320 MB/s, does not necessarily mean that the drive can actually burst at that speed. SCSI leaves extra bandwidth for other drives, remember, up to 15 drives can be connected to one channel. So in a multi-hard drive SCSI system, that overhead can easily get filled, but will not all be used with just one drive attached.

Share this post


Link to post
Share on other sites

Yep, if you have the Ultra and not the Ultra-64, you have the 33MHz/33bit bus. I was looking for the Ultra-64 myself but it was discontinued fairly quickly. Probably costs too much for most and it didn't sell.

I ended up with the Abit WI-1P which has PCI-X slots, although it's 66MHz--which is still not full speed PCI-X. I had to get a separate PCI-X SCSI controller, but the limit should be double of what you're seeing. In essence, my "dual controller" can really do one controller worth of drives.

Technically speaking, Gigabyte isn't lying. It really is an U320 controller. It's just not capable of U320 speeds.

Share this post


Link to post
Share on other sites

Yup, you are absolutely right, although that is irrelavant to the topic. For people who are getting confused here, the despute is over why he is not receiving 320 MB/s burst from one SCSI drive. Yes it is an Ultra 320 SCSI controller, yes it has up to 320 MB/s bandwidth, but the thing people need to realize is that with the current technology offered, no single SCSI drive will ever burst that high. An Ultra 320 SCSI controller was designed with that much bandwidth not for a single drive, but was designed with up to 15 drives in mind. That is what differs with SCSI vs. IDE. IDE channels are designed primarily for one drive's bandwidth needs in mind, please do realize though, and this too is a common misconception, that IDE drives do not, and will not (when you have 2 IDE drives connected to one channel) both access the controller/use bandwidth at the same time. IDE technology permits it from doing it otherwise. Where as with SCSI, all 15 drives connected to one channel can all be accessing data via the controller and using bandwidth. In which case, this is where 320 MB/s comes into play and is helpful. Because now what you do is, if all 15 drives are actually accessing the channel at the same time per say, then you would divide 320 MB/s by 15. (bandwidth / number of drives). That is why if you plan to use SCSI hard drives, just not in that amount, but only 2 SCSI hard drives, an Ultra 160 controller will do you just fine, you won't be able to max the 160 MB/s until you have 2-4 drives (many factors, dependent on what SCSI drives being used, how old the technology of the drives are, speed, etc...). Theoretically you could run 2 Ultra 320 SCSI drives on an ATA-133 channel and not max the throughput. That is because both drives will never be accessing the channel at the same time and because the latest SCSI drives do not have a sustained data rate (drive to controller) over about 80-90 MB/s, and do not have a burst more than 120-130 MB/s. Obviously there is more to that than just being able to work, because now you don't have the superior caching of a SCSI controller, drives can not simultaneously use the channel, and that is just in theory, so as you can see there would be many issues to be worked out first, that was just an example of bandwidth usages.

Anyway, I hope this clarifies it some.

Share this post


Link to post
Share on other sites

There is nothing that would theoretically stop a single drive from bursting at ridiculously high speed.... for something already in its 8mb of cache. The trick is only in finding a benchmark stupid enough to measure this useless metric.

Burst=peak instantaneous speed, NOT sustained speed.

Share this post


Link to post
Share on other sites

Another thing to consider is that the cache is only 8MB, which is small compared to the drive size and the size of files used during benchmarks. Exactly like bfg9000 said, the problem would be finding a benchmark capable of measuring the cache stream for that brief second (ms' for that matter) before the cache is not good anymore and needs to be refreshed, from that stand point, yes you could have a burst of 320 MB/s. The only problems would be detecting it in that split second it is cached, because most of the data would be random after that.

Share this post


Link to post
Share on other sites
Another thing to consider is that the cache is only 8MB, which is small compared to the drive size and the size of files used during benchmarks.  Exactly like bfg9000 said, the problem would be finding a benchmark capable of measuring the cache stream for that brief second (ms' for that matter) before the cache is not good anymore and needs to be refreshed, from that stand point, yes you could have a burst of 320 MB/s.  The only problems would be detecting it in that split second it is cached, because most of the data would be random after that.

Benchmarks that test the burst speed do so by repeatedly reading in the same few sectors on the hard drive. By doing so it ensures that the data is the drive's cache. This then becomes purely a test of the read speed of the drive's buffers, creating a test that should be a reasonable measure of burst speed.

Therefore it is reasonable to wonder why an U320 drive is only achieving 113.5MB/s, as the original poster has done. As has been determined already, in this case it is the PCI bus limitation.

Share this post


Link to post
Share on other sites
..and this too is a common misconception, that IDE drives do not, and will not (when you have 2 IDE drives connected to one channel) both access the controller/use bandwidth at the same time.  IDE technology permits it from doing it otherwise.  Where as with SCSI, all 15 drives connected to one channel can all be accessing data via the controller and using bandwidth.  ..<snip>..  Because now what you do is, if all 15 drives are actually accessing the channel at the same time per say, then you would divide 320 MB/s by 15.  (bandwidth / number of drives).

This is another common misconception. SCSI drives DO NOT access the bus at the same time. As with all bus technologies, there is only ever ONE device active on the bus at any moment in time.

Where SCSI technology is superior is it's efficient use of the bus. A SCSI controller sends commands to a drive, then the drive collects the requested data and readies it for transfer in the cache - it is not transferring data on the bus at this point. When it's ready to send the requested data to the controller (when the cache is filled or when all requests have been collected) it requests control of the bus, and when it gets it's turn it quickly sends the data out at burst speed.

Using this method of taking turns on the bus means all drives (up to the max. 15) can be filling their caches simultaneously and sending the data at full burst speeds in turn. As the number of drives increases, the time between each drive's turn on the bus also increases. Conversely, as the burst speed of the drive increases, the time required on the bus to send the data collected in the cache decreases. This can make the drive's capability to effectively burst at it's rated speed important.

The above also provides explanation as to why even an U160 SCSI controller is without a doubt sufficient to handle any of today's fastest U320 drives when used singularly - it has the bus to itself and can stroll along leisurely-like at the media's transfer rate of 70MB/s or so.

Share this post


Link to post
Share on other sites

The Gigabyte manual does spell out "320MB/sec" in their description of the U320 controllers.

This is irresponsible advertising. The manual does not state exactly how the U320 controllers are wired into the motherboard. Gigabyte should have gone on to state that the actual burst rates are limited by the saturation of the 33MHZ PCI Bus.

Do I have a good argument?

I want to pursuade them to refund me money so that I can replace the motherboard with one that would allow me to approach the ideal U320 burst rate or give me a voucher for a future Gigabyte Motherboard.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now