Self-healing disk arrays from Xiotech.com/Seagate

Roger Kelley from Xiotech presented to UUASC last week on Xiotech’s
next-gen storage technology. The basic ideas are:

– Xiotech is a spin-off from Seagate and still maintains an R&D
relationship with Seagate

– The basic disk storage unit from Xiotech is a unit that holds two
sealed-in datapacs, each of which holds 10 drives. so a minimum config
is 10 drives. max config for this single unit is 2 x 10 3.5″ drives or
2×20 2.5″ drives.

the max max config is a whole bunch of these units, going into 1
Petabyte. they call their device the “emprise”.

http://xiotech.com/Products-and-Services_ISE.aspx

– Xiotech & Seagate have 75 patents on new technologies that went into
this device. This is new stuff, folks.

– The basic storage unit, as far as the Xiotech RAID controller is
concerned, is not a disk or a partition but a platter. One drive has 4
platters and 8 heads. (Each platter has 2 sides.)

The monitoring is done using Seagate’s in-house technology, they get drive
telemetry continously and can take a platter side or a whole drive out of
service before it fails — sometimes a drive just needs to cool off and
then it can go back into production with no problem.

– Device interfaces to the outside world are Fibre channel or iSCSI over
1000 Mbps ethernet. Back end connectivity is Fibre Channel fabric which
eliminates Fibre Channel Arbitrated Loop and its associated problems.

– the Emprise is shipped with extra drives (4 extra drives – 2 per datapac)
locked inside the box.

The box does not open by the way! It’s not field-servicable!

The volume is striped at the head level. If there is a physical failure of
some portion of the drive (usually a head failure) the head is deactivated
and the capacity is replaced from the “reserve” area made up of the extra
drives, no field service is required, the device just keeps on ticking.

also, due to the hidden 4 drives, you can get 100% utilization on the device
(in terms of disk space) without any performance degradation since 16 drives
are striped across 20 drives. At 100% full on the logical side, the
physical drives are still only 80% filled.

In other words, there are eight drives per datapac as far as the customer
is concerned, but really there are 10 drives, because Xiotech adds 2 drives.

– RAID reconstruction does not kill you. Performance takes a dip but
not a dive since RAID reconstruction only occurs if there is data under
the bad head and only the affected head needs to be rebuilt. The other
heads which have no problems continue to provide data.

– Drives are positioned so that vibrations from rotation cancel each
other out. Vibration is the main source of drive failure and Xiotech
has substantially reduced it. They’ve also reduced heat, because at high
vibrational rates, additional piezoelectronics in modern drives activate
to compensate. Since those levels of vibration don’t occur heat is ALSO
reduced.

Statistics: Xiotech has sold 150 units since their product became available
in late June 2008 and have not had a single event requiring field service.

I find the “graceful degradation” engineering in this device beautiful and workable!

Xiotech has been testing in-house for 15 months, 208 units with 5900 drives. ZERO
field service events. And that’s using all drives returned to Seagate as BAD DRIVES.

Is that impressive or what?

The storage is sold at premium prices.