SerialATA WhitePaper
- Thread starter thorin
- Start date
The bottleneck isn't the bus, it's the speed of the drives. All SerialATA is going to do is remove some of the limitations of the bus, like the number of devices and simultaneous activity, and increase the maximum transfer rate. If you're pulling data from a single drive, you're still not going to get any better speed than your ATA66 drive unless it's buffered already or the manufacturers make SerialATA drives a huge bit faster.
<< The bottleneck isn't the bus, it's the speed of the drives >>
I don't think that's true. Granted, some 5400 RPM drives aren't all that fast in read times but I think the biggest limitation of the bus is it's parallel nature. By going serial you basically have a pseudo-SCSI design which is far superior to parallel ATA in its current form. If you look at that white paper you see that both drives on the serial ATA channel behaves like a "master" drive with a direct link to the ATA interface.
Also I think drives can read data at MUCH faster rates than say, 66MB/s (burst ATA/66 speed). I don't have time right now to do calculations, but given platter density and rotation speed, I'm POSITIVE that the final internal read rate of the drive is much much higher than 66MB/s sustained.
even w/o the parrallel limitation the drives still wouldnt behave much better. I have a friend w/ an adaptec 29160 scsi card and a Segate x15. He gets about ~45MB/s sustained rate... this is on a scsi u160 bus w/o the limitations of the ide bus. Say serial ata drives can get 45mb/s sustained, that is still far less than ata66 speed (66mb/s). Granted you cant put x15's on ide and/or you wouldnt want to... its just for example. te x15 is seen as the fastest scsi drive at this time and still cannot fill a single ata66 channel (not taking into account burst). As you can see then, the drives themselves have to get faster to take advantage of the rise in bus speeds. Granted, the physical ability to sustain transfer rates may be higher but the "actual" speed is what counts... not the theoretical.
sorry if i jumped from one place to another and back again... my point is that drive spindle speeds and latency needs to improve along w/ bus speeds in order to make a real difference. we can have 1GB/s bus but if our drives can only transfer ~35-45mb/s sustained maximum... it does us no good... well unless you start doing some raid 0 action...
sorry if i jumped from one place to another and back again... my point is that drive spindle speeds and latency needs to improve along w/ bus speeds in order to make a real difference. we can have 1GB/s bus but if our drives can only transfer ~35-45mb/s sustained maximum... it does us no good... well unless you start doing some raid 0 action...
If the read rate is more than 66MB sustained, why can't any current drive maintain a read rate of anywhere near even 66MBps? Because there is no drive that can maintain that speed. ATA66/100 is only useful for burst transfers, and even then many drives can't get that high. As far as the bandwidth, SerialATA will only increase the amount that's available to the drives for bursting, AND for their simultaneous operation. So combined throughput of multiple drives will need the higher bandwidth, but an individual drive simply will not benefit from it in any way that couldn't be duplicated with parallel ATA (because it's dependent on the drive hardware).
The serial nature of SerialATA will be an advancement that will allow better performance and more flexibility, however this is not some brand new mind blowing technology. They're just going the next step from what we've got now.
By the way, ATA100 already exists, SerialATA will only be 150MBps in 2002...do you really think UDMA isn't going to continue getting faster?
I admit that SerialATA is going to be in some ways better than current ATA, however it still won't be quite as good as SCSI (but the price difference will most likely make it a far better value), and it won't be a tremendous performance or technological leap over ATA or SCSI. It's just evolution, not a Big Bang. I just can't get excited about SerialATA for that reason, and because it's been in the works for a while and keeps getting press releases and stuff, but it's still at least a year away. That just annoys me.
The serial nature of SerialATA will be an advancement that will allow better performance and more flexibility, however this is not some brand new mind blowing technology. They're just going the next step from what we've got now.
By the way, ATA100 already exists, SerialATA will only be 150MBps in 2002...do you really think UDMA isn't going to continue getting faster?
I admit that SerialATA is going to be in some ways better than current ATA, however it still won't be quite as good as SCSI (but the price difference will most likely make it a far better value), and it won't be a tremendous performance or technological leap over ATA or SCSI. It's just evolution, not a Big Bang. I just can't get excited about SerialATA for that reason, and because it's been in the works for a while and keeps getting press releases and stuff, but it's still at least a year away. That just annoys me.
"The bottleneck isn't the bus, it's the speed of the drives"
I both agree and don't agree.
1) The drives today don't even challenge ATA66 (66mb/sec) let alone ATA100.
2) The bus is the bottleneck because it doesn't allow simultaneous access of multiple drives.
The best things about SerialATA aren't the bandwidth, but things like:
? Hot Plugability
? Reduced electrical requirements
? No jumpers
? Extremely small cables (check the diagrams the Data cable is smaller then the power cable....think about a day when ribbon cables to impede air flow, etc...)
? Backwards and future compatability
? Max 10% price premium vs todays drives (likely none)
(Yes the MAX internal transfer rates of current drives is 220mb/sec+ however you have to take into acount what this means. ie: the rate at which the drive can move data from the platter(s) to the internal cache or read buffer).
Thorin
I both agree and don't agree.
1) The drives today don't even challenge ATA66 (66mb/sec) let alone ATA100.
2) The bus is the bottleneck because it doesn't allow simultaneous access of multiple drives.
The best things about SerialATA aren't the bandwidth, but things like:
? Hot Plugability
? Reduced electrical requirements
? No jumpers
? Extremely small cables (check the diagrams the Data cable is smaller then the power cable....think about a day when ribbon cables to impede air flow, etc...)
? Backwards and future compatability
? Max 10% price premium vs todays drives (likely none)
(Yes the MAX internal transfer rates of current drives is 220mb/sec+ however you have to take into acount what this means. ie: the rate at which the drive can move data from the platter(s) to the internal cache or read buffer).
Thorin
<< I'm POSITIVE that the final internal read rate of the drive is much much higher than 66MB/s sustained. >>
Not sure where you get this from but the read rate of the drives is no where near 66MB/s. An IBM 75GXP can sustain ~37MB/s and it's the fastest ATA drive there is right now in terms of STR.
Lord Evermore, you're right SATA isn't a huge leap over PATA, but Hot Swappability and Small cables is pretty cool. It'll also be very nice to be able to drop the whole Master/Slave bit and just put your drives where you want them and not have to worry about concurrent access.
"Not sure where you get this from but the read rate of the drives is no where near 66MB/s. An IBM 75GXP can sustain ~37MB/s and it's the fastest ATA drive there is right now in terms of STR."
Read any drive spec, internal transfer rates are well over 220mb/sec.
Thorin
Read any drive spec, internal transfer rates are well over 220mb/sec.
Thorin
Ok so
Internal drive transfer rates are over 220MB/s
Externals are under 40MB/s
Why is that? It's not IDE's fault because SCSI isn't that much better.
Internal drive transfer rates are over 220MB/s
Externals are under 40MB/s
Why is that? It's not IDE's fault because SCSI isn't that much better.
It's exactly the same as the transfer rates between a CPU and L1 or L2 cache vs the transfer rate between CPU and any other system device.
Thorin
Thorin
Forgetting sustained read rates won't SATA increase burst rates? Right now a 75GXP drive has burst rates of 94MB/s evn though sustained rates are only around 37MB/s.
<< 220mb/sec >>
This is about a good seat-of-the-pants figure of internal transfer rates. The newer drives have ~300MB/s transfer rate. High end SCSI drives (15000 RPM) have > 580MB/s. I've done this calculation before but I can't find it... grrrrr...
<< Why is that? It's not IDE's fault because SCSI isn't that much better. >>
Try using the new U160 LVD SCSI cards. I believe you don't get 160MB/s transfer because of the limitation of the PCI bus bandwidth. But it is definitely possible with SCSI SSDs (solid state drive) because the drive simply saturates the entire SCSI bus.
SATA will increase both sustained and burst rates, with more emphasis on the sustained rate. Burst rates are important in PATA because the sustained rate is so low, you need the burst rate to average out the specs so that dumbass marketing can proclaim "Lightning fast 66MB/s transfers!".
I missed this earlier:
<< Say serial ata drives can get 45mb/s sustained, that is still far less than ata66 speed (66mb/s) >>
ATA/66 (66MB/s) is burst transfer, not sustained. You simply don't compare the two. Sustained ATA/66 is like 12MB/s at best.
<< It's exactly the same as the transfer rates between a CPU and L1 or L2 cache vs the transfer rate between CPU and any other system device. >>
Oh ok I thought it was the other way around for hard drives where buffer to host was extremely fast but the hard drive platter to buffer was the slow step.
"This is about a good seat-of-the-pants figure of internal transfer rates. The newer drives have ~300MB/s transfer rate. High end SCSI drives (15000 RPM) have > 580MB/s. I've done this
calculation before but I can't find it... grrrrr..."
Wel I remember when I was looking at KA and KX drives (Quantum) they were 237mb/sec, so with a small margin of error for my memory as it gets older I threw in 220
It may be closer to 300 by now, the point is that it's far above the Burst transfer rates of any current drive/interface.
Thorin
calculation before but I can't find it... grrrrr..."
Wel I remember when I was looking at KA and KX drives (Quantum) they were 237mb/sec, so with a small margin of error for my memory as it gets older I threw in 220
It may be closer to 300 by now, the point is that it's far above the Burst transfer rates of any current drive/interface.Thorin
Great post thorin.
<<By the way, ATA100 already exists, SerialATA will only be 150MBps in 2002...do you really think UDMA isn't going to continue getting faster?>>
Yes I don't believe UDMA will advance. According to the drive manufacturers, ATA100 was the LAST non serial arcitecture. Look at the SerialATA group, it's all the drive manufacturers. They are going to kill almost all SCSI with serialATA....
<<By the way, ATA100 already exists, SerialATA will only be 150MBps in 2002...do you really think UDMA isn't going to continue getting faster?>>
Yes I don't believe UDMA will advance. According to the drive manufacturers, ATA100 was the LAST non serial arcitecture. Look at the SerialATA group, it's all the drive manufacturers. They are going to kill almost all SCSI with serialATA....
I would think that SATA 100 would be slower then PATA100. Because wouldn't it face the same problem that everyones favority memory RAMBUS faces in that it has very high latenecy?
Uhmmm...assume say...240mbps internal transfer...that's 30MBYTES per second...I think someone is reading a data sheet wrong, and it's indicating megabits per second, when everybody else references in terms of megabytes per second.
Internal transfer rate is calculated based on spindle speed and density alone. The drive turns at a certain speed, and the transfer rate from that is far lower than the rate that the drive's buffers could accept the data or pass it out, and far less than the bus can handle. If the internal rate was faster than the external, then we would always max out the bus speed with every sustained transfer. The only time the bus speed is even approached is when the buffers already are full of data and they burst it out.
My Maxtor 30GB DiamondMax Plus 40 drives have average sequential transfer rates of 27MBps, maximum of 40MBps... They burst up to over 65MBps. (ATA100 drives and controller). HDTach can't even register higher than 80MBps, so I'm not sure what I was getting in RAID 0.
I do agree that the specs of SATA provide much greater features and flexibility, however everyone agrees that that part is much better; the argument is over whether there will be any performance difference. I also think that any parallel architecture is better than serial, and that the use of a serial design was only to allow them to advertise higher frequencies, similar to Rambus, where they can advertise an 800MHz bus, but it's transferring only a quarter of the data per cycle as a parallel architecture. This MAY allow higher frequencies in the future, however I don't believe that the serial design is just plain better as far as that goes. I'm sure that parallel could achieve just as high a transfer rate with the right designs.
Internal transfer rate is calculated based on spindle speed and density alone. The drive turns at a certain speed, and the transfer rate from that is far lower than the rate that the drive's buffers could accept the data or pass it out, and far less than the bus can handle. If the internal rate was faster than the external, then we would always max out the bus speed with every sustained transfer. The only time the bus speed is even approached is when the buffers already are full of data and they burst it out.
My Maxtor 30GB DiamondMax Plus 40 drives have average sequential transfer rates of 27MBps, maximum of 40MBps... They burst up to over 65MBps. (ATA100 drives and controller). HDTach can't even register higher than 80MBps, so I'm not sure what I was getting in RAID 0.
I do agree that the specs of SATA provide much greater features and flexibility, however everyone agrees that that part is much better; the argument is over whether there will be any performance difference. I also think that any parallel architecture is better than serial, and that the use of a serial design was only to allow them to advertise higher frequencies, similar to Rambus, where they can advertise an 800MHz bus, but it's transferring only a quarter of the data per cycle as a parallel architecture. This MAY allow higher frequencies in the future, however I don't believe that the serial design is just plain better as far as that goes. I'm sure that parallel could achieve just as high a transfer rate with the right designs.
Recneps:
Being serial is only a small part of Rambus's latency and that only becuase when you string more than one RIMM together it increases total trip time.
In this case serial just means it sends one bit at a time instead of lots through lots of wires. The actual protocol is the same as parrallel ATA they just send data 1 bit at a time instead of 30-40 (not sure how many pins in PATA are signals).
Being serial is only a small part of Rambus's latency and that only becuase when you string more than one RIMM together it increases total trip time.
In this case serial just means it sends one bit at a time instead of lots through lots of wires. The actual protocol is the same as parrallel ATA they just send data 1 bit at a time instead of 30-40 (not sure how many pins in PATA are signals).
"Uhmmm...assume say...240mbps internal transfer...that's 30MBYTES per second...I think someone is reading a data sheet wrong, and it's indicating megabits per second, when everybody else references in terms of megabytes per second."
Nope I meant MegaBytes per second.
Thorin
Nope I meant MegaBytes per second.
Thorin
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter