Quote:
|
and the average shard would go way past the average write cycle limit in a month or two on even the best flash storage...
|
Two years ago, I'd have agreed with that statement, but the write limitations of flash memory have gone up by a factor of 1000 since devices like USB drives became commonplace. They now utilize many of the same error checking methods as commercial SSDs. Essentially, they track the default bit-state by arranging each written bit in a way that it's two nearest neighbors will serve the dual funcion of parity checking. While this does require more space, modern flash usually has about 30% more room than the OS registers. So if the write limit is exceeded, it still knows the value of the bit that should be there. You would have to overwrite that 30% of the disk in excess of 100,000 - 300,000 times before data loss could occur.
Also, old flash memory used to read/write each bit from the same location. Newer memory is smarter. If there is free room on the drive (it requires about 15% free) it will attampt to move clusters of frequently accessed bits to infrequently used areas of the physical disk.
Also worth noting is the majority of changes that occur on a server can take place in RAM and changes don't necissicarily need to be written back to the physical disk, or are only saved periodically. This ofcouse depends on how the server is written. I haven't even looked at the RunUO Source, so I may be speaking beyond my level of experience. Basically, Servers built for speed with a large capactity for simutanious users tend to opporate in system memory most of the time. This means they take a ton of RAM and a dedicated server is usually necessisary. Servers complied for smaller users bases or lower system requirements would have frequent disk writes to ensure data integrity. I suspect RunUO is the latter, favoring security and and lower requirements over fast speed with high requirements. But that's just speculation on my part, if anyone more familiar with server source would like to correct me, I welcome their input.
Also, while I love my SCSI drive, a 10kRPM SATA II will out benchmark it unless your RAID them. While the SCSI has a 1.5x faster rotation speed, SATA has a 2x greater bit density. That means it
(potentially) reads twice as many bits per rotation, even while moving at 2/3 the rotational speed. This assumes bits are arranged contiguously. So a quality SATA drive should theoretically see about a 13% increase in speeds. SCSI however has othe driver controls allowing for on-the-fly defragmentation, error checking etc. SATA can do these things, but typically with greater system overhead. Both ofcourse require software to do this, but most packages are under $100 now, and there's always a torrent... For my money, it's a wash. They breakeven over standard usage, so save a few hundred bucks and go with SATA.