SSDs - State of the Product?
- Thread starter ddrueding
- Start date
ddrueding
Fixture
Just doing some photo work. My current Pano/HDR full 32-bit workflow involves PTGui, Photomatix, and CS4. Having one image in each stage and surfing while they are all processing keeps things busy.
ddrueding
Fixture
Do you like apples?
Mercutio
Fatwah on Western Digital
Firefox is the #1 tool for my work and for my productivity. I don't mind if it's using a lot of RAM, given the way I use it. Same thing with VMware and Thunderbird.
Thunderbird seems to take up the same 80MB RAM on all my Windows machines. I'm fine with that; there's 30,000 messages in my main inbox.
On the other hand, Winword.exe (2007) on my work destkop is using 67MB of RAM right now for two two-page-long documents. OpenOffice Write has similar levels of piggish behavior. Apparently that's just how modern word processing apps are.
Paint.NET, my image editor of choice, has 11 unsaved 1440x900 bitmaps (screenshots) in memory right now and is using a whole 14MB of RAM. I realize it can't do many of the thing you people are using Photoshop for, but on the other hand I'm also utterly convinced that Adobe can't code its way out of a web paper bag.
Mercutio
Fatwah on Western Digital
My commit charge is 1564MB at the moment out of a total of 4956MB available. That's with 300MB of Firefox and a 512MB VMware session. Process Explorer says PaintDotNet has 97MB of private bytes, which I take to mean the combination of pagefile, system libraries and RAM claimed by the executable.
64GB X25-E is dropping. Hmm.
ddrueding
Fixture
ddrueding
Fixture
I am enjoying my RAID-0 of 2x 32GB X25-Es. If you want to get the 64, I'll take the 32 off your hands
ddrueding
Fixture
I haven't had any time to check. Just let me know what you would like to see and I'll do it.
Mercutio
Fatwah on Western Digital
Obviously you are overclocking. What kind of RAM are you using for that setup?
My new gaming system is going to be an i7 920 sitting on a Gigabyte motherobard and three X25-Ms, probably in RAID0.
Some intel SSD's may use an update:
http://www.hardocp.com/news.html?news=Mzg5NjEsLCxoZW50aHVzaWFzdCwsLDE=
http://www.intel.com/support/ssdc/index_update.htm
http://www.hardocp.com/news.html?news=Mzg5NjEsLCxoZW50aHVzaWFzdCwsLDE=
http://www.intel.com/support/ssdc/index_update.htm
ddrueding
Fixture
Will this do? Windows dynamic disk striped using defaults.
ddrueding
Fixture
Santilli
Hairy Aussie
- Joined
- Jan 27, 2002
- Messages
- 5,278
David:
The default cluster size is 4kb, right? that would mean 71 mb/sec and 91 mb/sec? Does the cluster size limit the data transfer rate to just that one figure?
I will say I wished you hadn't put up that ram disk test. Your machine is now more then twice as fast as mine, bus speed, and, I think that indicates processor wise. That's my level for considering upgrades, or a new setup.
The default cluster size is 4kb, right? that would mean 71 mb/sec and 91 mb/sec? Does the cluster size limit the data transfer rate to just that one figure?
I will say I wished you hadn't put up that ram disk test. Your machine is now more then twice as fast as mine, bus speed, and, I think that indicates processor wise. That's my level for considering upgrades, or a new setup.
ddrueding
Fixture
I don't understand what you are saying here?
I suspect that it is much more than twice as fast. Keep in mind that I'm keeping my RAM clocked at 1066Mhz to keep it stable at faster timings.
Santilli
Hairy Aussie
- Joined
- Jan 27, 2002
- Messages
- 5,278
http://www.ntfs.com/ntfs_optimization.htm
http://support.microsoft.com/default.aspx?scid=kb;EN-US;q314878
http://support.microsoft.com/default.aspx?scid=kb;en-us;Q174619
After reading this, I still wonder what the best cluster size is for your raids?
If the cluster size is set at 4k, does that mean that you are limited to the speed of the 4k writes and reads on ATTO?
http://support.microsoft.com/default.aspx?scid=kb;EN-US;q314878
http://support.microsoft.com/default.aspx?scid=kb;en-us;Q174619
After reading this, I still wonder what the best cluster size is for your raids?
If the cluster size is set at 4k, does that mean that you are limited to the speed of the 4k writes and reads on ATTO?
ddrueding
Fixture
I've just left them all at defaults. I was mucking about for a while, and didn't see any that stood out (>10% difference). Even the Raptors can manage 250MB/s...so I think my storage is up to the task. Most of my photo editing stuff is now CPU locked again (stupid lack of multithreading).
I can't say specificly about ATTO, in that a benchmark program may be designed to specificly bypass normal Windows operation.
However, the general answer is no. Both Windows and Vista try to batch service I/O calls into sequential 64K blocks. It can be lower (for non sequential data or infrequent drive accesses) and I've seen 128K block sizes for raided volumes too but the vast majority of reads/writes are in 64K blocks. So as long as the disk isn't highly fragmented or it is dealing with small files scattered all across the drive (forcing a large proportion of seeks) the cluster size is not the determining limiter for drive speed. Further, as long as the data transfer is sequential (like in drive to different drive file copies), the head will not move between block read/writes so cluster size becomes even less meaningless if the files are big.
An important use of this data is when doing a file copy to and from the same disk. You can roughly estimate how long the copy will take (assuming no fragmentation). Since you know the drive will do 64KB blocks and the pattern is seek, read, seek, write you can calculate (number of 64K blocks) * (2 * ave seek time + Block size/ read sequential transfer rate + block size/ write sequential transfer rate).
So as an example using slow desktop drive copying a 1GB file, 15.9 ms drive with a read transfer rate of 60MB/s and a write transfer rate of 80MB/s produces 1GB/64KB * (2* .0159s +64KB/60MB/s + 64KB/80MB/s) = 526 seconds or 8.77 minutes. Now lets compare with a fast SCSI drive (Cheetah) 1GB file, 3.9ms average seek, 125MB/s read and 100MB/s write) = 1GB/64KB * (2 * .0039s + 64KB/125MB/s + 64KB/100MB/s) = 139.9 seconds or 2.33 minutes. And a fast SSD (X25-E) with 1GB file, .1ms seek, 230MB/s read and 200MB/s write will do the copy in 1GB/64KB * (2 * .0001s + 64KB/230MB/s + 64KB/200MB/s) = 12.5 Seconds
Slow desktop drive = 526 Seconds
Cheetah SCSI drive = 139.9 Seconds
X25-E SSD = 12.5 Seconds
Just as a side thing I just decided to do similar calculations with ddrueding's airplane simulator which would read from lots of random 4K files (i.e. 4K block reads between seeks with no writing) Rather than how long it takes to copy 1GB I'll report how many bytes/second which is more pertainant to the game.
Desktop 15.9ms average seek, 80MB/s read. So to read 4KB will take .0159s + 4KB/80MB/s = .01595 seconds or 256KB/s
4 raid0 desktop drives 15.9ms average seek, 320MB/s read. so a 4KB read will take .0159s + 4KB/320MB/s = .01591 seconds or 257KB/s (totally equivilent to a single drive)
Cheetah 3.9ms average seek, 125MB/s read so to read 4KB will take .0039s + 4KB/125MB/s = 1.041MB/s (4x faster than desktop drive)
X25-E .1ms seek, 230MB/s read: So to read 4KB will take .0001s + 4KB/230MB/s = .00017 s or 35MB/s (136x faster than a 4 drive raid0 desktop drive array)
ddrueding,
Now you know why I recomended an SSD for the application rather than raided desktop drives. The application is totally seek limited and sequential transfer rate is meaningless. Even the worst SSD is going to be far better than the best hard drive: The only thing that matters is the seek time and the outright capacity to store all those files.
Desktop 15.9ms average seek, 80MB/s read. So to read 4KB will take .0159s + 4KB/80MB/s = .01595 seconds or 256KB/s
4 raid0 desktop drives 15.9ms average seek, 320MB/s read. so a 4KB read will take .0159s + 4KB/320MB/s = .01591 seconds or 257KB/s (totally equivilent to a single drive)
Cheetah 3.9ms average seek, 125MB/s read so to read 4KB will take .0039s + 4KB/125MB/s = 1.041MB/s (4x faster than desktop drive)
X25-E .1ms seek, 230MB/s read: So to read 4KB will take .0001s + 4KB/230MB/s = .00017 s or 35MB/s (136x faster than a 4 drive raid0 desktop drive array)
ddrueding,
Now you know why I recomended an SSD for the application rather than raided desktop drives. The application is totally seek limited and sequential transfer rate is meaningless. Even the worst SSD is going to be far better than the best hard drive: The only thing that matters is the seek time and the outright capacity to store all those files.
Santilli
Hairy Aussie
- Joined
- Jan 27, 2002
- Messages
- 5,278
Kind of ironic that after 10 years my original argument, placing access time over SDT is being clearly shown to be accurate, thanks to SSD's.
The one thing you forgot to factor in is that the Raided SCSI drives do have the advantage of so far superior interface, and caching. Given time, or even I wonder using the SCSI interface, SSD's can have both the better caching system, better interface, and fantastic
access time.
Not to mention when you compare even the 3.9 ms access time, the SSD is nearly 400 times faster...
The one thing you forgot to factor in is that the Raided SCSI drives do have the advantage of so far superior interface, and caching. Given time, or even I wonder using the SCSI interface, SSD's can have both the better caching system, better interface, and fantastic
access time.
Not to mention when you compare even the 3.9 ms access time, the SSD is nearly 400 times faster...
Try 40x (not 400x) faster 3.9 vs .1
Also, after ATA-33, the interfaces have always been much faster than the drives, so the interface was not really a limiter. Drive-level caching really won't help when doing either of the above two senareo's because it is just plain way too small.
Really drive level caching only helps if the data set is very limited (like in a FAT table or a directory), where the drive is accessing the same data repeatedly, or where the data if very localized so as to prevent rotational latency.
Your 10Y arguement promoting average seek times over raw sequential transfer times has been true since the start. There are really a limited subset of drive activities (the main one that I can think of is a file copy between two independant drives) that actually can use raw sequential data transfer times to their benefit. That's why raid-0 and the newer drives data density has had a very limited value to me especially considering the extra data risk for Raid-0.
Really drive level caching only helps if the data set is very limited (like in a FAT table or a directory), where the drive is accessing the same data repeatedly, or where the data if very localized so as to prevent rotational latency.
Your 10Y arguement promoting average seek times over raw sequential transfer times has been true since the start. There are really a limited subset of drive activities (the main one that I can think of is a file copy between two independant drives) that actually can use raw sequential data transfer times to their benefit. That's why raid-0 and the newer drives data density has had a very limited value to me especially considering the extra data risk for Raid-0.
Santilli
Hairy Aussie
- Joined
- Jan 27, 2002
- Messages
- 5,278
Thanks.
For what it's worth, going from a 60 mb sec to 105 mb/sec Seagate SATA drive produced a noticeable change in OS response, and, a noticeable change and snap in loading. After that, the difference diminishes.
Considering the failure rates of current SATA/PATA drives, one really wonders if the best way to go isn't a refurbished SCSI drive for a boot drive? I'm currently using a 10k, 147 gig Mat boot drive, in the game machine. It's snappy, and considerably faster then a SATA boot drive in the same machine.
Random Acess is 7.6 ms
Sustained is around 90 mb/sec. While not a 15k, it's an economical alternative, and, they can be had for 60-150 dollars in SCA.
While I'm not denying the benefits of the SSD's, they have a long way to go, both in price and interface, at least from what we've noted so far.
For what it's worth, going from a 60 mb sec to 105 mb/sec Seagate SATA drive produced a noticeable change in OS response, and, a noticeable change and snap in loading. After that, the difference diminishes.
Considering the failure rates of current SATA/PATA drives, one really wonders if the best way to go isn't a refurbished SCSI drive for a boot drive? I'm currently using a 10k, 147 gig Mat boot drive, in the game machine. It's snappy, and considerably faster then a SATA boot drive in the same machine.
Random Acess is 7.6 ms
Sustained is around 90 mb/sec. While not a 15k, it's an economical alternative, and, they can be had for 60-150 dollars in SCA.
While I'm not denying the benefits of the SSD's, they have a long way to go, both in price and interface, at least from what we've noted so far.
Stereodude
Not really a
That's some expensive RAM. 4GB of CAS4 DDR2-800 is under $50.
Both!
There are good free programs for XP32. The best one I found for XP64 is a cheap shareware.
Mercutio
Fatwah on Western Digital
Well, with three X25-Ms in RAID0, my City of Heroes level load times are about 1/6th what they are on a 7200rpm magnetic disk. Which is pretty huge.
On the other hand, for $1200 worth of hard drives, I'd damned well better get 5 second load times.
My i7/920 is presently sitting at 3.6GHz and I'm on stock-everything. I have pretty much the least ambitious RAM I could've purchased for overclocking adventures.
On the other hand, for $1200 worth of hard drives, I'd damned well better get 5 second load times.
My i7/920 is presently sitting at 3.6GHz and I'm on stock-everything. I have pretty much the least ambitious RAM I could've purchased for overclocking adventures.
Mercutio
Fatwah on Western Digital
Well, with three X25-Ms in RAID0, my City of Heroes level load times are about 1/6th what they are on a 7200rpm magnetic disk. Which is pretty huge.
On the other hand, for $1200 worth of hard drives, I'd damned well better get 5 second load times.
My i7/920 is presently sitting at 3.6GHz and I'm on stock-everything. I have pretty much the least ambitious RAM I could've purchased for overclocking adventures.
On the other hand, for $1200 worth of hard drives, I'd damned well better get 5 second load times.
My i7/920 is presently sitting at 3.6GHz and I'm on stock-everything. I have pretty much the least ambitious RAM I could've purchased for overclocking adventures.
Santilli
Hairy Aussie
- Joined
- Jan 27, 2002
- Messages
- 5,278
That's what's listed for my motherboard. Not sure what I can go with that would work, and be faster.
I'm considering just leaving it alone, and, buying a new motherboard, ram, etc. The i7, and the accompanying bus speed looks like a pretty good jump. IIRC, the Athlon 3200 is really running at 1.8 ghz. The i7 Mercutio is using would fit my at least 100% faster before I replace policy.
I'm considering just leaving it alone, and, buying a new motherboard, ram, etc. The i7, and the accompanying bus speed looks like a pretty good jump. IIRC, the Athlon 3200 is really running at 1.8 ghz. The i7 Mercutio is using would fit my at least 100% faster before I replace policy.
Mercutio
Fatwah on Western Digital
One thing you need to understand, Greg, is that not all clock cycles are created equal. You can't just say "twice as fast" and have it be so. Newer Intel chips have shorter execution pipelines and literally do less work to perform computation, which is part of the reason AMD is getting spanked so hard by Intel right now.
Besides that, these are four core systems, which is an advantage all its own.
sechs
Storage? I am Storage!
I dunno. After having four cores for so long, two seems pretty pokey to me.
Santilli
Hairy Aussie
- Joined
- Jan 27, 2002
- Messages
- 5,278
I've always liked duals, at least. STILL don't find twin 2.8 ghz Xeons anywhere near slow.
However, having a nearly 9 year old machine...
On the otherhand, when SSD's finally get down far enough in price, dragging SCSI with them, that old dual machine might just get a new lease on life.
Now that this Athlon is a HTPC, I should probably just build a new box for games...
Mercutio: Odd that you would post that. The entire Athlon numbering system for those processors was designed to indicate how their slower processors would compete against the faster Intel processors. Intel Xeons, despite sometimes slower clock speeds, would often be VERY fast, despite lower numbers then some of the other Intel processors.
Sleepy...
However, having a nearly 9 year old machine...
On the otherhand, when SSD's finally get down far enough in price, dragging SCSI with them, that old dual machine might just get a new lease on life.
Now that this Athlon is a HTPC, I should probably just build a new box for games...
Mercutio: Odd that you would post that. The entire Athlon numbering system for those processors was designed to indicate how their slower processors would compete against the faster Intel processors. Intel Xeons, despite sometimes slower clock speeds, would often be VERY fast, despite lower numbers then some of the other Intel processors.
Sleepy...
timwhit
Hairy Aussie
Those Athlon numbers were made to compare to the P4. After the Core2Duos came out those numbers were useless for comparison purposes. If you read about the architecture of the P4 versus the newer Intel chips you'll understand this better.