Richard Scobie wrote: > If one disregards the rotational speed and access time advantage that > SAS drives have over SATA, does the SAS interface offer any performance > advantage? It's a very good question, to which I wish I have an answer myself ;) Since I never tried actual SAS controllers with SAS drives, I'll reply from ol'good SCSI vs SATA perspective. > For example, assume a SAS drive and a SATA drive can both sustained > stream 70MB/s. A 16 drive JBOD SAS enclosure with internal SAS expander > is connected via a 4port SAS RAID controller, configured for RAID 5 > across all 16 drives. > > If tests are then run reading and writing a multi gigabyte file to empty > arrays made up of 16 SAS drives and 16 SATA drives, would the results be > identical? In this scenario, I don't think there will be any difference between the two. With SCSI, one is limited by SCSI bus speed, which is 320Mb/sec (640Mb/sec for two-channels etc). Hopefully there isn't such a limit in SATA/SAS world, at least not in theoretical configuration. In your case, 4port SAS controller is about 1200Mb/sec theoretical maximum (300Mb/sec per port), which is further limited by system bus speed (PCIe or PCI-X). But note that while you're doing an operation in one stream, you'll most likely be limited by a speed of a SINGLE drive, since at any given time, you're writing/reading only to/from one drive (I'm simplifying things here, -- omitting calculating/writing parity info in raid5, which may be further limited by controller's XOR operation speed -- let's assume it's a raid0 array for now, not raid5). The only case where more than one drive is affected is when you cross strips - go from one drive to another in a raid layout. When a controller receives a request which spans strips, it can split the request into two and issue parallel requests to two drives. Here, we will see speed 2x larger than of a single drive... Where things starts to become different is when there are MULTIPLE readers and writers, so that an I/O subsystem has much more chances for parallelism. And here, it all depends on the controller, and probably on the driver too, AND on the drives. Here are several observation I collected in several years of expirience with various raid controllers and drives. First of all, I've yet to see a controller that is really able to handle multiple requests in parallel. Usually, multiple I/O threads gets exactly the same summary performance as a single thread - UNLIKE of linux software raid which clearly scales exactly to the number of drives (having in mind raid0 again and omitting raid5 complexity). All raid cards from Adaptec (at least up to and including their first SATA cards), from HP (SmartArrays), from LSI shows this behavior - no matter how many drives you throw at them, no matter how many threads you start, the result will be the same. Many external "smart" disk arrays (not JBOD boxes but with RAID controllers inside, connected using FC or iSCSI, with built-in stacability and external redundancy) also show this behavior. Maybe, just maybe, I've only dealt with their "cheap" products so far, but I lost any hopes long ago, and decided that HW raid solutions is something to avoid as much as possible. Now for the actual difference between SATA and SCSI drives. And here, again, is something interesting. They says that modern SATA drives has NCQ, which is "more advanced" than ol'good TCQ used in SCSI (and SAS) drives. I've no idea what's "advanced" in it, except of that it just does not work. There's almost no difference with NCQ turned on or off, and in many cases turning NCQ ON actually REDUCES performance. It's in a very good contrast with SCSI drives, where - depending on workload (I'm again talking about multiple I/O processes - for single I/O there's no difference in NCQ/TCQ or raid arrays/controllers), you may get 20.. 400% increase in speed - of a SINGLE drive. Even 9Gb SCSI disks from Seagate - circa 1998 - shows this behavior (they're alot slower than todays ones, but speed increase due to TCQ is very well of the same extent). With linux software RAID configuration and with TCQ-enabled SCSI drives, there's a very significant difference between SATA and SCSI. Linux raid code can do things in parallel, and individual SCSI drives plays here nicely too, so the end result is very good. With SATA, linux raid still plays its role as with SCSI, but the drives don't scale at all. And with hardware raid controller, even the raid code doesnt scale - so end result is like single drive in single-stream test, which is the worst of all. I'd LOVE to be proven wrong, but so far the only evidence I see is the one that proves my observations. Recently I removed an IBM SeveRaid module from a rackmount IBM server (had to find how to disable the damn thing and expose AIC79xx controller - it turned out to be quite difficult) and replaced the raid with linux software raid solution - server speed (it's a database server) increased about two times, so that users noticied it immediately (hey, our server works unusually fast today!). Go figure. > I ask, as I have seen a comment to the effect that SATA drives are less > efficient interacting at the bus/interface level in this situation, but > I have had no luck confirming this after extensive searching. I recall: single-stream tests will not show any measurable differences. Its the multi-reader-writer tests which will. /mjt - To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
