SKIP HOFMANN wrote:
the FASTEST NAS is BLUEARC
???
Hello Skip
Do you have real new comparative test results or are you just citing the press releases from Bluearc? ;-) http://www.bluearc.com/html/news/press_releases/pr_specsfs_110104.shtml
Neither NetApp nor BlueArc are the fastest NAS providers ... They are both "slow" compared to ... Exanet, IBM, Panasas, Spinnaker, ...
Let's have a look at the NFS-Benchmarks BlueArc cites ( http://www.spec.org/sfs97r1/results/sfs97r1.html ) I will focus on the TCP results:
Company System Proto Result Overall Response Time BlueArc Inc. Titan 32 TCP 101571 1.78 Exanet Inc. ExaStore EX600FC TCP 178156 1.04 IBM Corporation IBM eServer p5 570 TCP 145362 1.17 Panasas, Inc. Panasas ActiveScale TCP 305805 1.76 (!!!) Spinnaker Networks SpinServer 4100 TCP 134385 1.90
But ... first let's stay at a small BlueArc <-> NetApp comparison as this is the toasters list. :-)
BlueArc Titan 32 50858 1.76 Many io/s with bad response time. NetApp FAS980 36063 0.95 Less io/s with fast response time.
The Fas980 has three quarters (72 %) of io/s with almost half response time (54%) in this benchmark. (Yes, I know about the saturation with high loads... More about this later.)
Does the fastest mean 1) highest i/o rates? 2) lowest response times? 3) (1) and (2) at the same time? 4) With which kind of protocols? UDP / TCP, NFS / CIFS / ...?
If you check the details, you can see many of them getting into saturation/overload problems....
http://www.spec.org/sfs97r1/results/res2003q4/sfs97r1-20031103-00166.html The graph really shows that Panasas wanted to hit the 300k mark and hit the saturation with 7.8 msec. :-)
Same thing with BlueArc: http://www.spec.org/sfs97r1/results/res2004q3/sfs97r1-20040913-00203.html NetApp already stops at 2.5 msecs and not at 5.2msecs response time like BlueArc does ...
One of the lousiest results comes from the EMC Celerra NS600 cluster: 38459 ios @ 10.3 msecs ... ( http://www.spec.org/sfs97r1/results/res2003q1/sfs97r1-20030203-00125.html )
Let me try to have a look at still comparable "Real" and not Overall Response Times at 35k-40k io/s: (That's where NetApp stops with the single head.)
Vendor Model CPU RAM NVRAM IO/s "Real" RT @ 35k-40k GB GB (if value is available, if not, next higher value) BlueArc Titan 32 1(+7) 14+8 1 40488 2.6 (1 FS) BlueArc Titan 32 2FS-Mod 2(+14) 28+8 2 40513 1.3 (2 FS) Exanet EX600FC 12 72 84 36390 0.8 IBM eServer p5 570 4(8) 128 16 44175 0.7 NetApp FAS980 2 8 0.5 36063 2.5 (1 FS) NetApp FAS980c 4 16 1 41157 1.0 (2 FS) Panasas Active Scale 60 240 240 60864 0.8 Spinnaker SpinServer 4100 12 24 10 35536 1.1
Smaller (not even the half) RAM size than Titan 32: ==================================================== The FAS980 (8 GB RAM) is still almost as fast as the Titan 32 (22 GB):
36063 @ 2.5 msec. <-> 40488 @ 2.6 msec.
Same RAM size comparison: ==================================== The FAS980 cluster (16 GB RAM) is 2.5 (!) times faster than the Titan 32 (14+8 GB): 41157 @ 1.0 msec. <-> 40488 @ 2.6 msec.
Dual "Head" comparison: ===================================== The FAS980 cluster (16 GB RAM) is still faster than the Titan 32 with two FileSystemModules (24+8 GB) which is not even is a cluster. 41157 @ 1.0 msec. <-> 40513 @ 1.3 msec.
b.t.w. :The FAS980c is a real cluster and afaik BlueArc doesn't have clustered high availability solutions: http://www.bluearc.com/html/library/downloads/titan_spec.pdf
=> If I compare the RAM and cache sizes of NetApp to the competitors ... I finaly see one thing: NetApp offers a very thin, io-optimized operating system with an excellent filesystem that is able to create a still "good" performance with even small "cheap" hardware. Please, don't start discussing the prices for the software licenses... ;-) but afaik they are still acceptable. :-)
So, if you change your expression about the fastest BlueArc NAS-solution to something like "fastest NAS with one single CPU (with many (7) FPGAs support), with ...". In this very small area, BlueArc "appears" to have the fastest solutions, currently. Imho: If NetApp would start the same lousy saturation trick with the FAS980, they would reach the same results. But beside this small area where BlueArc "tops", I only can see big marketing efforts of them in trying to cheat/fool/mislead (don't know which one of these expressions is the softest...) the customers in yelling out loud "we are the fastest". And this is something I really don't like ... and reminds me to other storage vendors I also prefer to not use ... ;-)
So imho we should give the "Fastest-NAS-crown" to the IBM eServer p5 570 (SingleHead/Backplane) 145362 @ 1.17 or the Panasas AS Cluster (ClusteredMultiHead) 305805 @ 1.76. BlueArc gets my current "We did the lousy saturation trick and/but yell it out loud" crown. :-( ;-)
So therefore my question once again: Why do you believe that "Bluearc" is the fastest NAS-solution provider? I tried to explain my current opinion and hope you or anyone else can enlighten me if I'm wrong.
Best regards! Dirk Schmiedt
By the way: I still haven't found out, wether the SFS97r1 response time is measured on the client (including network delays) or on the server side? Can anybody explain this to me?
So therefore I'm still dreaming of a multiprotocol high available, fault tolerant clustered NAS head which can do takeovers without loosing the states/connections (even at CIFS), with OS upgrades without visible reboots, with ... and it should not cost more than 1000$ @ head ... Just dreaming ... ;-)