Video Capture Directories and USB Flash Drives

[BonzaiDuck]

I've been experimenting with VISTA-64, and installed a Patriot USB "Ready-Boost"-capable flash drive on one dual-core machine. After convincing myself of the noticeable performance improvement from Ready-Boost, I decided I wanted to try a flash-drive as destination for video-captures on an XP-MCE-2005 system.

After changing the capture-software's target/destination directory to the flash-drive (an ample 32-GB Patriot XPorter), the video and audio begins to stutter and freeze, with square-shaped artifacts in the video.

I then discovered that the partitions on flash-drives are default to FAT-32, and that there is a way to make them NTFS partitions -- with forum-posts and other sources showing great performance improvements when flash-drive formats are converted to NTFS.

But I still have the same symptoms with video capture.

Video playback -- with MPEG2 files copied to the flash drive -- is just fine.

The system with XP has a Creative Labs PCI audio card, and a single (BFG) nVidia 8800 GTS graphics card. The capture-card is a PCI_E Avermedia Combo M780.

Can anyone illuminate for me why my expectations for using a large flash-drive in this way are unreasonable? I cannot find any IRQ conflicts in the system, but I can look again.

 

[BonzaiDuck]

ADDENDUM: Could it be that the format "block-size" may change the performance of a flash-drive? The default under FAT32 is 512 bytes. The maximum is 64K.

 

[mpilchfamily]

Your slow up is probably in the transfer speed of the flash drive or even the speed at which the USB control is able to pull the information. Flash drives actually perfom slower then a typical HDD. Now an SSD does have much faster read performance then most HDD and deffinetly faster then a conventinal flash drive.

 

[BonzaiDuck]

I must have been misled by unwarranted assumptions as to why a "ReadyBoost" flash-drive improves overall performance.

Those assumptions may be reflected in the difference between "write" (capture) and "read" (play) performance per MPEG2 files.

I'm also guessing that the performance of ReadyBoost depends more on reads than writes.

Oddly, though, I recall a blog with tables of performance results where sustained read and write speeds improved by factors exceeding between 10 and 100 when drive formats were changed from FAT32 to NTFS. The MB/sec results for a flash-drive would still indicate better performance than a single hard disk. Maybe I mistook "mega-bits" per second as "mega-bytes" per second, but the data were described as MB/s.

I'm also wondering whether the cache-less, bufferless configuration of a flash drive may also have something to do with the lackluster performance.

With the Ready-Boost feature, you'd think the advantage combines (1) lower power consumption of the device itself, (2) removing the problem where a hard-disk is simultaneously writing and accessing a swap-file while it may be writing to the disk, and (3) the read performance as opposed to write performance.

Well -- if the newly-announced Intel X-25M SSD only has a starting price of $600, it will be interesting to see how that price trends over the next year or so.

 

[Denithor]

Originally posted by: BonzaiDuck
I'm also guessing that the performance of ReadyBoost depends more on reads than writes.

Exactly. Which is why most of the current crop of MLC drives have the poor multitasking performance they exhibit. (The Intel SSD being the obvious exception, thanks to their proprietary controller.)

Well -- if the newly-announced Intel X-25M SSD only has a starting price of $600, it will be interesting to see how that price trends over the next year or so.

I think we're all looking forward to this with great interest. If the intel drives were priced around $300 I would probably buy one right now, but $600 is just too hard to swallow. I mean, I can build a whole system for that kind of cash.

 

[VirtualLarry]

Yeah, USB flash drive simply have horrible write performance, most of them. For video capture, you need decent write performance.

The reason that USB drives work for readyboost, is because of seek time, they are negligible, not because of bandwidth.