RE: IDE controller card and Fedora Core 3

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



James' analysis is right on the money, *BUT* for a given file set,
Robert is also correct, and I've proven it with some equipment of mine.
The maximum transfer speed to/from a given machine never goes above a
certain point no matter 100Mbs, Gbit, Ultra, Ultra2, Ultra3 - doesn't
matter.  32bit PCI is the likely bottleneck these days.

Having said that however, RAID *IS* a great way to squeeze performace
out of older hardware for file storage and sharing purposes.
Distributing reads/writes through a raid controller enables utilization
of that maximum bandwidth, which would not otherwise be possible through
a single drive - even newer drives.

Paul

> -----Original Message-----
> From: fedora-list-bounces@xxxxxxxxxx 
> [mailto:fedora-list-bounces@xxxxxxxxxx] On Behalf Of James Wilkinson
> Sent: Monday, December 06, 2004 6:35 AM
> To: For users of Fedora Core releases
> Subject: Re: IDE controller card and Fedora Core 3
> 
> Robert L Cochran wrote:
> > You want to RAID5 640 Gb  worth of drives on a Pentium II, 400 MHz 
> > machine? I'm not a RAID expert. I question whether doing it 
> on an old, 
> > slow machine with large drives is practical. It seems to me 
> you would 
> > simply saturate the PCI bus.
> 
> Depends. (And I know the Original Poster later said performance wasn't
> critical...)
> 
> Firstly, unlike certain OSes, Linux should find that a 
> Pentium II is fine for shoving data around. It's not like it 
> needs to do complex calculations on each byte: it shouldn't 
> even need to examine each byte while reading.
> 
> Of course, while it's writing, something will have to calculate
> checksums: I don't think the Original Poster's suggested card 
> has any RAID acceleration features.
> 
> Secondly, it depends exactly what you're going to put on that 
> RAID array. If you're putting a mail server or a database on 
> it, you'll almost certainly find that disk usage is dominated 
> by lots and lots of small (~8K) reads. And you'll be very 
> lucky to get more than about 150 of those per second on ATA drives. 
> 
> It's all about physics: on a hard disk that spins about 100 
> times per second, a particular byte only goes near the read 
> heads once every 10 ms. And no matter how fast the rest of 
> the computer is, 10 ms is an age in computing. Database 
> administrators swear by the "number of spindles", for this 
> reason: I've got much chunkier disk arrays on not much faster 
> systems. (They also swear *at* RAID 5 for a lot of usages...)
> 
> And 8K times 150 transfers per second = 1200 K/s per disk: 
> that's not going to set anything on fire.
> 
> (The Original Poster mentioned SAMBA, but it still depends on 
> what sort of files are being stored).
> 
> Thirdly, on a slow PCI implementation, you're going to get 90 
> - 100 MB/s of usable bandwidth. Hard drives are only just 
> going past the 70 MB/s at peak performance. At the best of 
> times, you are going to be limited by the PCI bus, but you're 
> still going to get better than a single drive on the latest systems.
> 
> Of course, on a Samba server, you're going to be sharing that 
> bandwidth with the network card (on a PII 400). If that's 
> gigabit, then yes, that's going to halve the available 
> bandwidth and slow things down. If it's only 100 Mbit/s, then 
> that's 12.5 MB/s max, and that's your
> bottleneck: that's the physical maximum you'll get out of the system.
> 
> (When did 100 Mbit/s become "only"?)
> 
> But this is why PCI Express is a Good Thing. It's designed 
> for this sort of job: a separate network card and a separate 
> fast hard drive adapter.
> The bandwidth isn't shared: they each get their own 2.5 
> Gbit/s per lane in each direction.  Although you might 
> *still* need more than 1x PCIe for the hard drive adapter...
> 
> James.
> -- 
> E-mail address: james | Anonymous:          What do you think 
> of Stainer's
> @westexe.demon.co.uk  |                     "Crucifixion"?
>                       | Sir Thomas Beecham: Good idea! 
> 
> --
> fedora-list mailing list
> fedora-list@xxxxxxxxxx
> To unsubscribe: http://www.redhat.com/mailman/listinfo/fedora-list
> 


********************************************************************************
This message is intended only for the use of the Addressee and may
contain information that is PRIVILEGED and CONFIDENTIAL.

If you are not the intended recipient, you are hereby notified that any
dissemination of this communication is strictly prohibited. If you have
received this communication in error, please erase all copies of the
message and its attachments and notify Space Imaging immediately.
********************************************************************************



[Index of Archives]     [Current Fedora Users]     [Fedora Desktop]     [Fedora SELinux]     [Yosemite News]     [Yosemite Photos]     [KDE Users]     [Fedora Tools]     [Fedora Docs]

  Powered by Linux