PDA

View Full Version : Performance Tweaks



Bobt
05-04-2004, 02:47 PM
I have an IDE raid on a PIV 2.8. (Never could afford a SCSI)
I am trying to get the Max out of my VT3.
So, what settings should I be intrested in and do
you have any recomendations for a good setting.
What controls the number of layers you can achieve
etc.

Thanks,
Bob

rbartlett
05-04-2004, 03:29 PM
I'd be more specific if I could be certain about what is in that large cube of yours Bob. Is it i860 with a 3ware Escalade 7000 series?

If the IDE is on a card that is significantly parallel (3ware, RAIDCore) then you initially start off with the controller's proprietary striping mechanism. This isn't the last word on the matter. However I recommend the use of the vendor's solution where you want data redundancy and striping (as NT doesn't support both together and higher end storage handling)

If you have multiple generic controllers use NT/OS striping.

Essentially NT/OS striping is the basic denominator. You won't be far off the optimal and generally speaking the provided driver software for RAID will tend to be slower on the more generic "enterprise server" solutions. Due to the fact that the RAID is designed to deliver the high bitrates and low access times in bursts. Daftly this is more true of SCSI than IDE - such is the attraction for controller vendors to develop the SATA/IDE killer design.

3ware solutions have proven interesting when working with larger numbers of drives. Split the drives into two bunches of proprietarily controller striped volumes and then NT/OS stripe these. Take a guess at the best ratio split as it may depend.

Lastly, if you are in the mood to experiment, try these alternatives:

1. Don't stripe, if your sources are predictable and you don't tend to be breaking up a single original file, then this works. Multicam might be a good example, but you might be surprised how often a long run occurs on a layer without edits. It depends on the type of business you are in.

2. Leave the drives as separate volumes. Perhaps leave them with their device enumerations but unmapped to drive letters. Then use mountvol to put them into the C:\ device space. Saves having to remember the drive letters you've allocated to each purpose. (the disk space on C:\ only ever reports the space left on C:, and not the sum of C: with the sub-ordinates, not much more than that)

NB, you need to be good at housekeeping to enjoy this version. Every VT-Edit stream has its own directory. Works if this matches your workflow. This method also mitigates IDE's random access/latency hit compared with 10k/15k SCSI.

3. RAID-10, improving random access time and providing top class performance striping. You might also care for the redundancy this offers. Should the implementation be good from the proprietary driver that supports it. YMMV

Assumptions:

That the IDE controller/s are on a separate bus to the VT PCI card (or even separate from the DV OHCI function if switching at the DV level). i875P-6300ESB, i860, E7505 and E7520 are remarkable for how they assist with this ambition in a good number of their implementations.

Pete Draves
05-05-2004, 09:31 AM
I usually set the bus (32 or 64) that the controller is on to max transfer rate
This usually keeps any problems away
Pete Draves
System Integrator
for Taylor Viseo