Sean Cummins on Mon, 13 Jan 2003 22:21:06 -0500


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

[PLUG] Re: Server Recommendation


gabriel rosenkoetter writes:

Well, in all fairness, the FCAL arrays we've got are mid-range at
best. They're Just a Bunch of Disks, really. The truly sexy stuff,
that you get from Hitachi or EMC, does *way* more in "hard"ware
(firmware, or even software, really; the EMC Symmetrics array has
its own fairly large OS development team, for instance... they're
not developing the interface for OSes to use, but the OS that runs
on the array itself), making full mirrors of RAID 5 arrays synchronized
at all times (at no performance loss), which you can then split
off from the set, then backup to tape from that. Or, if you're
*really* spendy, you've got a full set (or two) locally, and another
full set on the other end of some dark fiber (which you either run
fiber channel or a regular networking protocol down) and do off-site
vaulting, all the rage if you're a data warehouser.

And to add to this (I have to; I work for EMC PS [<- that was my disclaimer]), the Symmetrix can even use SRDF to replicate data over IP, so we have customers who are using it for cross-country disaster recovery. Typically people will have local "3rd mirror" (aka BCV) copies of their data internal to their primary Symm(s), split the BCVs off at some point during the day to create point-in-time snapshots of their production data, and then incrementally copy the data from the BCVs over some kind of WAN link (essentially transport independent, but there are latency requirements) to remote Symm(s). And all of this happens automatically (with scripting), without the host OSes being aware of it going on in the background. This allows organizations to virtually guarantee data integrity -- its how many companies in downtown NYC recovered from 9/11. You can combine SRDF (symm-to-symm replication) and TimeFinder (replication internal to a single Symm) in an incredible number of ways, like multi-hop replication (Symm -> Symm -> Symm -> etc).


As for some specs, the current (but not for long) top-of-the-line Symm, the 8830, can support ~70TB of raw space in a 3-bay cabinet (3 floor tiles), 64GB of cache, eighty 333mhz PowerPC procs, and something around 350 hosts on a SAN (although I've never seen anyone put that many hosts on a single Symm). One of our larger customers (who has around 300 Symms and at least a few Petabytes of storage capacity -- we have a fulltime dedicated staff there [I'm not one of them]) actually uses the standard RAID 10 (striping across mirrored volumes, as opposed to RAID 0+1 mirroring across striped volumes) internal to their Symms, and uses host based LVM to stripe across the RAID 10 volumes, effectively creating huge RAID 100 logical volumes. Its pretty interesting stuff.. </salespitch> :)

- Sean

_________________________________________________________________________
Philadelphia Linux Users Group        --       http://www.phillylinux.org
Announcements - http://lists.netisland.net/mailman/listinfo/plug-announce
General Discussion  --   http://lists.netisland.net/mailman/listinfo/plug