Back!
After a highly implausible RAID1 (mirror) stripe failure on Thursday, I have been mostly blasted off the net. Both disks failed within minutes of each other, with physical defects in similar areas of the disk. 10 out of 10 for consistency for Samsung (they made these disks), they were same make, same model, same batch and consecutive serial numbers. Quite infuriating, though.
So, while the disks are on their way back to the manufacturer (only 7 months old with a 3 year warranty), I had to attempt a lateral approach. All I had was lemons, so it was time to start making lemonade. All my servers were still healthy, including the big ph4t Skynet. So I thought, OK, maybe I can network boot it. The first thought that came to mind was: "Boromir, that's retarded." But it just so happens that I upgraded my network to gigabit last week, so the result may not actually end up being wholly unusable.
I've managed to get my box up and running with PXE+NFSRoot, running off good old trusty Skynet. If truth be told, I cannot actually perceive any difference in performance. If anything, running it off a NFS root might actually be slightly faster. I guess I upgraded to gigabit [puts little finger on lower lip and engage in maniacal laughter] ethernet just in time. :-) Although it adds 0.1ms latency, it benefits from rootfs server's caching which defeats the 8ms latency of the disks, resulting in an substantial overall win! Plus, the 6 disks in Skynet between them are easily capable of saturating a gigabit link, something that Raiden's 2 disks probably couldn't have done despite being much, much newer (6 years or so, but still same RPM, 4x the size and 2x the cache per disk).
So I am quite tempted to actually keep things running as they are. It also means that if this machine blows up entirely, I could just plug another one in and it would just pick up where the old one left off. Of course, the downside is that if Skynet were to go belly up, it'd be doubly bad (which is why I'm paranoidly checking that it's 6-disk RAID5 stripe is happy quite often at the moment).
So: "Did you see that?! Did you fscking see that?! Two hits with one shot! Owned! Fscking OWNED!" :-D
So, while the disks are on their way back to the manufacturer (only 7 months old with a 3 year warranty), I had to attempt a lateral approach. All I had was lemons, so it was time to start making lemonade. All my servers were still healthy, including the big ph4t Skynet. So I thought, OK, maybe I can network boot it. The first thought that came to mind was: "Boromir, that's retarded." But it just so happens that I upgraded my network to gigabit last week, so the result may not actually end up being wholly unusable.
I've managed to get my box up and running with PXE+NFSRoot, running off good old trusty Skynet. If truth be told, I cannot actually perceive any difference in performance. If anything, running it off a NFS root might actually be slightly faster. I guess I upgraded to gigabit [puts little finger on lower lip and engage in maniacal laughter] ethernet just in time. :-) Although it adds 0.1ms latency, it benefits from rootfs server's caching which defeats the 8ms latency of the disks, resulting in an substantial overall win! Plus, the 6 disks in Skynet between them are easily capable of saturating a gigabit link, something that Raiden's 2 disks probably couldn't have done despite being much, much newer (6 years or so, but still same RPM, 4x the size and 2x the cache per disk).
So I am quite tempted to actually keep things running as they are. It also means that if this machine blows up entirely, I could just plug another one in and it would just pick up where the old one left off. Of course, the downside is that if Skynet were to go belly up, it'd be doubly bad (which is why I'm paranoidly checking that it's 6-disk RAID5 stripe is happy quite often at the moment).
So: "Did you see that?! Did you fscking see that?! Two hits with one shot! Owned! Fscking OWNED!" :-D