DELL|EMC CX300 + PowerEdge 2850 + RHEL ES 3.0 U7 + REDHAT CLUSTERMANAGER SUITE 3.0
RankinK at orau.gov
Wed Apr 26 13:59:35 CDT 2006
It seems that the FAS3020 would be a better comparison.
I attached a screenshot from my camera phone that I took over the
shoulder of a technician the other day. It's of the remote view of
Windows XP running on the CX300. It would hit about 65% CPU utilization
with a single stream, and then it would peg 100% when another host used
a LUN on it's other storage processor.
Myself, I don't know how they can sell these devices in good conscience.
They advertise them as a high-performance device capable of handling I/O
up to over 600MBps, and advertise 29-64MBps of I/O to each SATA disk.
To me, they seem like they're better suited to be Fisher-Price's new
line of "My First Fibre Array" devices. We're currently in the process
of Dell trying to convince us that we don't need anything more than
20MBps of I/O for each of these: an 800 user Oracle Financials system,
600+ users of Windows DFS and a pre-(dual)LTO3 diskbackup solution.
That has honestly been the best that they have been able to sum up. In
9 weeks, and multiple onsite engineers, they've come to the point where
we have over 10 contacts to deal with, and soon, a group of Dell
employees will be onsite so they can "prove" that we don't need more
than 60MBps of sustained I/O out of two arrays that cost us a quarter of
a million dollars.
My recommendation would be to take a look at the Engenio devices (IBM,
Sun, Storagetek, SGI, Teradata, etc. all use them). They have hardware
XOR engines, don't use up your cache memory for their operating system,
and certainly don't behave like what I've seen out of these (When we
pull a controller for failover tests, it takes 10-15 minutes for "SOME"
of the LUNs to come back online). Storagetek's FLX380
(http://www.storagetek.com/products/product_page2702.html) is a 4Gbps FC
solution, supports up to 224 disks, has 4GB of cache, 8 fibre channel
paths and is still MUCH cheaper than these EMC CX300's. Their FLX280's
(could be compared to an IBM DS4300 Turbo) are a good deal cheaper than
the FLX380, too. In fact, we recently received the wrong quote from
IBM, and wound up with not a DS4300, nor a DS4300 Turbo, nor even a
DS4400. They sent us a quote for a DS4500 with 6 controllers, three
trays of fibre disks (that were twice as large as what we have; 140GB
disks whereas we bought 70GB disks), four trays of fibre disks, and the
wrong host connect kit, and a pair of them came out $5,000 more than
what we paid for two CX300. Basically, for 2% more, we could have had
two generations more powerful fibre array head, four more controllers,
twice as large fibre disks (and three times as many of them; 3 trays),
and twice as many SATA disks (an extra two trays, for a total of 4).
Not to mention you don't get handed to a company that doesn't go around
making a giant mess of Linux. I recently toured IBM's Linux labs in
Austin, TX, and was amazingly impressed by the people there, whereas
with my Dell/Linux experience, every time I turn around, I'm handed a
catch-22 problem locking you into specific old versions of software and
UNIX Systems Administrator
Information Systems Department
Oak Ridge Associated Universities
From: Irwan Hadi [mailto:ihblist at gmail.com]
Sent: Wednesday, April 26, 2006 2:06 AM
To: Rankin, Kent
Cc: linux-poweredge at dell.com
Subject: Re: DELL|EMC CX300 + PowerEdge 2850 + RHEL ES 3.0 U7 + REDHAT
CLUSTERMANAGER SUITE 3.0
Last year I had a performance issue with a Dell PE2800 with PERC 4e/di
running Microsoft Windows Server 2003 and has Gold support for our
Exchange 2003 server. It took around 2 months to get resolved, with
*NO* effort at all from Dell gold support. Took me around 200 hours
personal time during this ordeal, mainly friday and saturday night.
In the end my conclusion was a Dell PERC can not handle a RAID 5
volume with bigger than 4 X 146 GB hard drives or the random read
performance will be so poor that a tortoise is still faster.
The problem is not mainly with Dell though. It is with the
manufacturer of the PERC controller (LSI Logic in this case), as I
tried their MegaRaid 320-4X (rebranded as PERC 4/QC) which is supposed
to be LSI's top of the line product, and it performs even more poor.
Anyway, back to this EMC CX300 thing which supposedly to have
relatively the same performance with NetApp FAS270, the sequential
write seems to be correct. Our NetApp FAS270 can do sequential write
at about 60 MB/s before the processor is maxed. The thing that used up
the processor is mainly the parity calculation (either RAID 5 for EMC,
or RAID 4 / RAID DP with NetApp). Of course snapshot with NetApp only
takes < 1 second to finish, and doesn't need a special volume (BCV)
In any case, I strongly believe that every storage system either it is
from big vendors such as EMC, NetApp, Hitachi HDS, HP EVA, or small
vendors such as Equallogic, leftHand, bluearc, pillar data, etc. have
their own quirks, and also strengths. Problem it is, is to find and
work around the quirks.
So far we've been very happy and satisfied with NetApp, though it
costs about an arm (not yet an arm and a leg like EMC). We are not
that happy with LSI (due to their MegaRAID / PERC controller). We are
not quite happy with EMC as we've heard quite a few horror stories
On 4/25/06, Rankin, Kent <RankinK at orau.gov> wrote:
> Expect terrible performance, and even worse support.
> My organization is nine weeks into a support call regarding
performance problems on a pair of CX300's. When we initiate a snapshot,
we can't get more than 4-5MBps of sustained write performance out of a
3-tray CX300. When everything is just dandy, our processors max out
when any combination of LUN's is writing ~55MBps.
> The support situation has been an absolute nightmare, as Dell's sales
and support staff have been doing their best to discredit anything that
we say, rather than to address the problem at hand. For instance, we
bought one tray of SATA disks for a Windows fileserver, and can't get
more than 28MBps of sustained write to 14 500GB disks in RAID 5
configuration. They sold it for a Windows fileserver, yet claim that
Windows file copies are not an appropriate way to claim whether or not
it functions to spec. *sigh* I won't even get into the fact that they
claim that another tray of SATA disks are performing excellently for a
pre tape (2x LTO3 tape drive) disk backup solution, even though they
can't break 30MBps.
> In my experience, so far, the best systems to look at are the Engenio
devices sold by Storagetek, IBM, Sun, SGI, etc. Netapp also has devices
that will grossly outperform these. These CX300's have nice management
software, but are absolutely the worst performing devices that I've ever
> Not to mention that they sell them with 2GB of cache, and then use
about 1100MB of that to run Windows XP Embedded on the controllers. So
you wind up with a giant, gaping security hole (unpatched Windows), plus
under half of the cache that they advertise.
> Kent Rankin
> UNIX Systems Administrator
> Information Systems Department
> Oak Ridge Associated Universities
> From: linux-poweredge-bounces at dell.com
[mailto:linux-poweredge-bounces at dell.com]
> Sent: Tuesday, April 25, 2006 4:50 AM
> To: linux-poweredge at dell.com
> Subject: DELL|EMC CX300 + PowerEdge 2850 + RHEL ES 3.0 U7 + REDHAT
CLUSTERMANAGER SUITE 3.0
> Hello here DELL people,
> Good day!
> I just want to ask ifthere is any white paper or maybe a linkon DELL
that has implemented this configuration. I wish to setup a 2
DELLPowerEdge 2850 on aHA linux cluster using Redhat Cluster Manager
Suite 3.0 that I can use as a reference.
> Please help
> Linux-PowerEdge mailing list
> Linux-PowerEdge at dell.com
> Please read the FAQ at http://lists.us.dell.com/faq
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 29394 bytes
Url : http://lists.us.dell.com/pipermail/linux-poweredge/attachments/20060426/6d4be21b/attachment-0001.jpg
More information about the Linux-PowerEdge