Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
Wut?Buy SAS if you are awesome and need dual-port high availability features. SAS can read and write simultaneously, whereas SATA cannot. SAS can also handle higher QD loads much better.
SAS disks have WWNs for unique identifiers. Dual-port SAS disks have two of them. The enclosure will map one WWN to each port, and allow the two HBAs to talk to the one disk simultaneously. A SATA SSD doesn't use WWNs, so only one controller could talk to it at a time. I suppose it's possible to have a chip that converts SATA to SAS, and I think LSI even makes them, but I'm not familiar enough with them to tell you which features would or wouldn't work.TJ, nice post. I'm still trying to wrap my head around this stuff. So if I have an SAS JBOD enclosure with two host ports, and connect a different HBA to each port, are the HBAs talking to the separate phys on each disk? And if I put, say, a SATA SSD in the JBOD, is it even possible for both of the HBAs to talk to it?
I don't suppose you've got a link to a paper talking about how that works, do you? I think it might make for some interesting reading. I could see it working with multiple disks, but then again, I've not seen anything to suggest that SATA disks, even with the half-duplex limitation, wouldn't be able to do the same. I think I need to wrap my brain around it somehow. I'll gladly correct one person. If two people tell me I'm wrong, well... Maybe I'm not as right as I think.SAS can read and write at the same time. Again, we are talking about SAS, the protocol to talk to disks, how the disk does it does not matter.
what about a SAS SSD, slow too?
Those aren't fast at all by SSD standards. The Hitachi SSDs (which are SAS) that I have in my desktop for example are more than 2 years old and have far better sequential speeds and write IOPS. Endurance is considerably better too.http://www.sandisk.com/enterprise/sas-ssd/optimus-max-ssd/
Quite fast actually. I can't wait until those come out
That depends on the physical cabling, if I'm understanding you correctly. For example, you can have one SFF-8088 connected to a controller, and another connected to a backplane. (Via a 8088-to-8087 cable) You could also have one connected to the backplane's "IN" port, and one connected to the "OUT" port. Or, you could have both connected to the backplane's "IN" ports. Only the last scenario is true "dual port".My question is: if you have a two port jbod enclosure, do the two ports automatically talk to the different SAS ports?
Speaking of which, have you guys analyzed the kind of writes that a system drive undergoes on Windows Server? That is, if it's been configured as a discrete drive for Server with data on the array? I'm wondering what kind of feasible life the average enterprise MLC has because the prospect of applying updates and rebooting a server in like 30 seconds is appealing as all get outSAS and SATA are much more similar than IDE and SCSI ever were though. For instance, SAS controllers are compatible with SATA drives. The reverse (using a SAS drive on a SATA controller) is not possible however.
You often see situations where people use SATA SSDs instead of SAS drives in their servers because they are compatible, faster, and often cheaper.
I'm wondering what kind of feasible life the average enterprise MLC
Wow! I get it now about the provisioning, that totally makes sense about the size of the drive. About the backups, what is a good low maintenance strategy? RAID-1 would affect the life of the mirror just as much as the primary so is there some kind of delayed or periodic swappable drive software or hardware solution? Say for example the mirror drive is written to four times a day or so and only incrementally so it's not mirroring the entire thing? I guess just periodic images of the system drive would be an option but I'm thinking something that can come online quick like BackupExec Pro's imaging features.I expect several decades of life provided you get a 250GB SSD or larger. Although with that said any storage device can die at any time so always have a backup.
This. Set something up in Windows scheduler to do a robocopy /mir (mirror) once a day between drive 1 and drive 2. Or, in linux/unix/bsd/etc, add an rsync command to cron. You won't get real* backups that way, but it's basically RAID-1 without constant writes.RAID1 is fine for uptime but I would do a real backup instead or in addition to the raid.
Awesome!! Thanks guys!This. Set something up in Windows scheduler to do a robocopy /mir (mirror) once a day between drive 1 and drive 2. Or, in linux/unix/bsd/etc, add an rsync command to cron. You won't get real* backups that way, but it's basically RAID-1 without constant writes.
*Differential backups, backups set to read-only once they're written, etc
nearline vs online.One difference I noticed is SAS drives don't spin up until they receive the command to do so. In my case when Windows loads.
The Optimus SSDs are optimized for storage size, and can sustain "only" 1 to 3 full drive writes per day. The Lightning SSDs are smaller but twice faster, and can sustain 25 full drive writes per day. Still, it is amazing that they are able to pack 4 terabytes into a 2.5 inch form factor. That would be 16 terabytes if they had 5.25 inch SSD drives! It's rare that SAS drives are leading the edge for storage capacity compared to consumer SATA drives.http://www.sandisk.com/enterprise/sas-ssd/optimus-max-ssd/
Quite fast actually. I can't wait until those come out
This. Set something up in Windows scheduler to do a robocopy /mir (mirror) once a day between drive 1 and drive 2. Or, in linux/unix/bsd/etc, add an rsync command to cron. You won't get real* backups that way, but it's basically RAID-1 without constant writes.
*Differential backups, backups set to read-only once they're written, etc
Please note that this does not mean a drive will fail on average every 285 years. It means that if you had 2.5 million drives, one would fail per hour. Big, big difference.By the way, both the top Optimus and Lightning drives are guaranteed for 5 years, which matches the best enterprise hard drives, but their Mean Time Before Failure is 2.5 million hours, which if my computation is correct is about 285 years...
Please note that this does not mean a drive will fail on average every 285 years. It means that if you had 2.5 million drives, one would fail per hour. Big, big difference.