Door Anoniem: 4Terra nou nou daar gaan heel wat illegaal binnengehaalde filmpjes op ....

Door Anoniem: Helaas geeft die tabel onvoldoende data weer voor een eerlijke beoordeling. De 12 TB van Seagate scoort hier zeer slecht ondanks dat die disken nog zeer jong zijn. Zijn deze disken werkelijk onbetrouwbaar, of is de belasting gewoon veel zwaarder (%disk time)?

Shard Integrity Checks

The shard integrity check utility runs as a utility task on each Storage Pod. In late June, we decided to increase the rate of the shard integrity checks across the data farm to cause the checks to run as often as possible on a given drive while still maintaining the drive’s performance. We increased the frequency of the shard integrity checks to account for the growing number of larger-capacity drives that had been deployed recently.

The Consequences for Drive Stats

Once we write data to a disk, that section of disk remains untouched until the data is read by the user, the data is read by the shard integrity check process to recompute the hash, or the data is deleted and written over. As a consequence, there are no updates regarding that section of disk sent to SMART stats until one of those three actions occur. By speeding up the frequency of the shard integrity checks on a disk, the disk is read more often. Errors discovered during the read operation of the shard integrity check utility are captured by the appropriate SMART attributes. Putting together the pieces, a problem that would have been discovered in the future—under our previous shard integrity check cadence—would now be captured by the SMART stats when the process reads that section of disk today.

By increasing the shard integrity check rate, we potentially moved failures that were going to be found in the future into Q3. While discovering potential problems earlier is a good thing, it is possible that the hard drive failures recorded in Q3 could then be artificially high as future failures were dragged forward into the quarter. Given that our Annualized Failure Rate calculation is based on Drive Days and Drive Failures, potentially moving up some number of failures into Q3 could cause an artificial spike in the Q3 Annualized Failure Rates. This is what we will be monitoring over the coming quarters.

There are a couple of things to note as we consider the effect of the accelerated shard integrity checks on the Q3 data for Drive Stats:

The number of drive failures over the lifetime of a given drive model should not increase. At best we just moved the failures around a bit.
It is possible that the shard integrity checks did nothing to increase the number of drive failures that occurred in Q3. The quarterly failure rates didn’t vary wildly from previous quarters, but we didn’t feel comfortable publishing them at this time given the discussion above.