Actually, Mark Bentley's results were not in that thread I linked to earlier. Here's his later post with the results: http://teaparty.mathworks.com:1999/toasters/9944.html
Joe
-----Original Message----- From: Joe Luchtenberg Sent: Wednesday, January 30, 2002 1:37 PM To: 'Sam Schorr'; Mr. G. D. Geen; toasters Subject: RE: IP 4700 vs Netapp
Celerra is most certainly not the forerunner of the IP4700 (although it did precede the "Chameliion"). Celerra is a network attached front end EMC offers to Symmetrix customers to provide a NAS interface to the SYM. The IP4700 is a purely NAS device built on a h/w platform from the DG acquisition married to a NAS OS developed from the CrosStor acquisition. Both NAS products are current EMC offerings.
As to Mr. Geen's points:
- the 50K ops/sec if I understand correctly is a measure of
the memory to network mapping performance (remember the BlueArc threads of several months ago); you will find this performance statistic in EMC literature; I believe EMC did in fact improve *system* performance, and if memory serves me they went from roughly 5000 ops/sec in v1.0 to now 11,000 in v2.0 (for comparisons bear in mind the IP4700 is a two-node cluster); similarly for MB/sec statistic.
- disk management is also a point of comparison; IP4700 RAID
sets can only be 5- or 9-disks in size; since it is RAID 5, dynamic volume expansion is problematic (as you know) when compared to RAID 4.
- snapshot manageability is poor on the IP4700; I believe the
snap reserve space is hard-reserved, and in order to change this setting you must first delete all existing snapshots.
- the write-cache-to-disk on power outage is a nice feature.
Regarding performance Mark Bentley did some independent testing on an IP4700 vs. F840 that he discussed on Toasters last fall. The thread starts here: http://teaparty.mathworks.com:1999/toasters/9889.html ...and I believe he posted results later in the thread.
BTW, on a filer if you start with a RAID set consisting of 4x 36GB disks (3 data/1 parity) and add 4x 72 GB disks the resulting RAID set goes something like this:
- first 72GB disk added becomes new parity disk; 36GB disk
previously used for parity becomes a 36GB data disk
- next 72GB disk is only used to 36GB capacity (34GB,
actually -- same as the other 36GB disks)
- any additional 72GB disks added after that (in this case
two more) are used to their full capacity
- net result in this case is 4x36GB data disks; 1x72(36)GB
parity; 1x72(36)GB data; 2x72GB data
Generally, I recommend to create a new RAID set for the 72GB drives. Granted, this will result in two parity disks. However, you gain the following advantages (among many others):
- greater reliability, due to smaller RAID sets reducing risk
of double-disk failure
- simplicity
- equivalent capacity to adding the 72GB drives in a single
mixed-capacity disk RAID set (net result of half-capacity data disk described above)
- less expensive than adding more 36GB disks
Joe
Joe Luchtenberg Dataline, Inc.
Email: joe.luchtenberg@data-line.com Phone: 757-457-0504 (direct line) 757-858-0600 (front desk) 757-285-1223 (mobile) Fax: 757-858-0606
Please visit us at www.data-line.com
-----Original Message----- From: Sam Schorr [mailto:sschorr@homestead-inc.com] Sent: Wednesday, January 30, 2002 12:42 PM To: Mr. G. D. Geen; toasters Subject: RE: IP 4700 vs Netapp
G.D.
Are you sure about the statement that the EMC does not allow NFS and CIFS on the same volume? When we evaluated the Celerra
product (which
I'm guessing is the forerunner of the IP4700), it certainly allowed for this. If I recall correctly, the underlying file system
was UFSx, and
the CIFS was layered on through a version of Samba running
on SCO UNIX
(the Celerra controllers).
--sam
-----Original Message----- From: Mr. G. D. Geen [mailto:geen@ti.com] Sent: Wednesday, January 30, 2002 9:03 AM To: toasters Subject: Re: IP 4700 vs Netapp
Good day all,
I would like to respond to this topic. Though we have no EMC in our design environment, EMC has made several trips to Texas
Instruments to
inform us on their product line. Back in June of 2001, EMC reported to us on the capabilities of the EMC IP4700 and please let me know if they have made updates to the sales pitch since then.
The IP4700 is said to perform 50K NFS ops/sec. Now this is good but one must note that this is using four channels which computes
to 12.5K NFS
ops/sec per channel. That is less performance per channel than a single F720 NetApp filer. Along the same lines, EMC rated throughput as 340MB/sec. Again this is using all four channels.
As for scallability the new NetApp F840 scales just as well -- 12TB today and 18TB in a new DOT release due this year. This is roughly equal to that of the IP4700 as stated in June of 2001.
The IP4700 supports RAID 0, 1, 0+1, and 5 whereas the
NetApp is RAID4
and WAFL file system. I will let you make the call there but I am happy with the performance of the WAFL system and RAID 4 of the NetApp.
The storage processor of the IP4700 consisted of two PIII Celerons. This may be upgraded by now.
As for disk usage, I found this to be an interesting concept but I will let you decide which is better. On the EMC all disks in a RAID group must be the same size, no different from NetApp. Upon disk failure, however, the EMC grabs the first available disk. Now if you have multiple size disks in your system, the spare disk *must* be of the largest size you have on your system. On the NetApp, the filer will grab the smallest spare disk of equal or larger size of the one that failed. Thus, if you have 36GB and 72GB drives on your NetApp and a 36GB drive fails, the NetApp will try to find a 36GB spare.
If one is
not found then a larger spare will be resized and used as a spare.
Another interesting point is EMC's spare disk designation. On an EMC a spare disk is always a spare disk. If a disk failure were to occur, the RAID group will rebuild using a spare disk. Once the failed disk is replaced, the RAID group is once again rebuilds on to the replaced disk. For each disk failure there are two rebuilds of the data.
During this
time a filer is most vulnerable to a second disk failure. There is also performance degradation during rebuilds. On the NetApp,
there is only
one rebuild of the data on to the spare disk. When the
faild disk is
replaced, it becomes the spare.
Some other features of the EMC are mirror root disk, battery backup, and write cache mirrored to disk on power outage. The battery backup of the EMC is not meant to keep the filer running indefinitely, just long enough to move the write cashe on to disk and shut down the filer gracefully. NetApp is using "motor cycle" battaries in F800 series filers to maintain the NVRAM. Though I have never run in
to a problem
where I lost the NVRAM contents due to dead battery, there is the remote possibility.
One other issue, if you are running in a mixed environment, EMC does not allow for both NFS and CIFS to be running on the same volume. Both, though, may be running on the same filer. NetApp does allow you to run both NFS and CIFS file systems on the same volume.
Just some thoughts.
-gdg
Rahul Kumar wrote:
Hi I am also interested in knowing for the same.... EMC was recently at our place and they claimed to be faster than netapp's like ops /sec etc etc. Their demo box is yet to come. If anyone has tested IP4700 let us know Thanks Rahul
-----Original Message----- From: Mike Ball [mailto:MBall@datalink.com] Sent: Tuesday, January 29, 2002 8:06 PM To: toasters Subject: IP 4700 vs Netapp
Has anyone tested a EMC IP 4700 in the last 6 months? EMC
claims they
have greatly improved the box in the last six months, but I
have not
been able to get any specifics as to what they have done. Thank you, Mike