XtremIO leads the market with up to 6 successive SSD failures per X-Brick and multiple simultaneous SSD failures across multiple X-Bricks (so long as each X-Brick is failing 1 SSD at a time). Out of >20,000 SSDs deployed, we have only had 6 SSD failures to date, so the odds of 2 simultaneous failures within the same X-Brick are astronomically small. Successive SSD failures are a much more likely failure event, especially with other AFA’s write hot spots (which XtremIO never has). The key question is what number of successive failures can competitive AFAs support? And for those concerned with the simultaneous failure risk within a single X-Brick, we are writing dual parity per X-Brick today and will productize the rebuild logic for simultaneous dual SSD failures in H1 (as a non-disruptive upgrade from 3.0). It is also worth noting that with spinning disks, you needed two parity drives not really for two drive failures, but for 1 drive failure, and a single sector error on any other remaining drive. On SSD’s, the drives themselves guard cells with correctable ECC and hence are not prone to a “bad sector”. Again, the real world risk of simultaneous SSD failure is remote- but the probability of successive drive failure in other AFAs is very high with hot spotting.
yansheng
235 消息
0
2014年10月20日 21:00
Raid 6 是具备两个盘坏掉数据，
RADI6磁盘阵列级别是在RAID5基础上为进一步加强数据保护而设计的一种全新RAID方式，换句话也可以说是一种扩展RAID5级别。
如果一块硬盘出现物理故障，RAID的会处于降级状态，但是仍然有容错功能；
如果第二块硬盘出现故障，逻辑盘中还剩下的两块硬盘就不再有容错功能。
如果阵列中有hotspare硬盘，出故障的硬盘上的数据会转移到hotspare硬盘上，自动进行重建，数据的条带化存储方式和原来的两块硬盘相同。
不知道你是否明白？
Roger_Wu
4K 消息
0
2014年10月20日 22:00
这句话就是这个意思吧：不仅能防止两块盘故障，还能防止一块盘故障+另一块扇区错误。老外写技术文章喜欢写得很细，其实就是两块盘失效。
zhaos2
643 消息
0
2014年10月20日 22:00
谢谢。我是从下面的论述看到的
XtremIO leads the market with up to 6 successive SSD failures per X-Brick and multiple simultaneous SSD failures across multiple X-Bricks (so long as each X-Brick is failing 1 SSD at a time). Out of >20,000 SSDs deployed, we have only had 6 SSD failures to date, so the odds of 2 simultaneous failures within the same X-Brick are astronomically small. Successive SSD failures are a much more likely failure event, especially with other AFA’s write hot spots (which XtremIO never has). The key question is what number of successive failures can competitive AFAs support? And for those concerned with the simultaneous failure risk within a single X-Brick, we are writing dual parity per X-Brick today and will productize the rebuild logic for simultaneous dual SSD failures in H1 (as a non-disruptive upgrade from 3.0). It is also worth noting that with spinning disks, you needed two parity drives not really for two drive failures, but for 1 drive failure, and a single sector error on any other remaining drive. On SSD’s, the drives themselves guard cells with correctable ECC and hence are not prone to a “bad sector”. Again, the real world risk of simultaneous SSD failure is remote- but the probability of successive drive failure in other AFAs is very high with hot spotting.
Roger_Wu
4K 消息
0
2014年10月20日 22:00
有真实案例吗？一直以来接受的信息都是RAID 6可以防止两块硬盘同时失效。故障和扇区损坏当然也算失效啊。
不过RAID 6不同厂商的标准会有一些不同，Intel公司的P＋Q双校验RAID 6、NetApp公司的双异或RAID 6（也称RAID-DP）、X-Code编码RAID 6、ZZS编码RAID 6、Park编码RAID 6、EVENODD编码RAID 6等等。会不会是这些厂商自己RAID 6的限制呢？方便的话说一下这个说法的出处最好......
P+Q RAID 6：
RAID-DP：
还有些较少见的RAID 6就不列举了...
zhaos2
643 消息
0
2014年10月20日 22:00
是not really，不是not only
主要还是想说明两块盘同时坏的可能性很小，而几块盘的连续失效（不是同时）的可能性很大。
Roger_Wu
4K 消息
0
2014年10月20日 23:00
好吧，那就是XtremIO自己的限制。XtremIO的X-Bricks也不是标准的RAID 6吧。
zhaos2
643 消息
0
2014年10月20日 23:00
谢谢！绝大多数人都是这么理解的。只是我今天才看到那段英文论述，想求证一下。目前看来是没有什么理论依据。
zhaos2
643 消息
0
2014年10月20日 23:00
这里的RAID6是对传统磁盘阵列的，XtremIO用的是XDP，与传统的RAID完全不一样。
yansheng
235 消息
0
2014年10月21日 23:00
不要纠结这个