> This message is in MIME format. Since your mail reader does not understand
this format, some or all of this message may not be legible.
Hi Saker-
Thanks for the lengthy reply. Yeah, I heard that earlier versions of the
Extreme OS were ³buggy², but I thought the problems were confined to their
implementation of BGP. Doesn¹t sound like that applies to you. Maybe these
issues have been fixed with their move over to Linux. Hopefully, for their
sake, and their customers, they have. All in all their products are quite
reputable, but Force 10 is definitely coming on strong these days.
I agree with you regarding Force 10. I am currently learning toward the
S50¹s. It is simply a great stackable. It¹s more expensive than the Extreme,
but cheaper than Cisco (assuming you have negotiated deep discounts off the
street price). I likely will be ordering 5 of them to get started, though
final pricing isn¹t locked in quite yet.
It sounds like you are very pretty happy with Isilon stuff. I am currently
doing an eval of it on-site at a client¹s using four (4) of the 3000 nodes +
one (1) accelerator. It¹s holding up quite well, but then again they have
far fewer nodes than does Zoic. I am mildly disappointed that the total
throughput of a single cluster node can¹t seem to get beyond megs a second
(sustained). It can peak there with little effort but sustained throughput
tends to hover around 35 megs a second. Having said that, I like the
clustered file-system approach. It¹s ability to intelligently load-balance
network and disk IO is a huge selling point. I think that the move to SAS
drives will help future performance. I am also curious if their decision to
use TCP/IP for the data channel (over Infiniband) plays a role. I was a
little shocked to hear that they opted for it as their high speed transport.
Sure it keeps costs down, but what about performance? I am wondering will
they ever come close to saturating the 10GigE front-end they are testing
now. It surely will be interesting to see.
Yeah, I know what you mean about SATA drives. MTBF ratings are a ruse if you
ask me. Spindle speed and count matter the most, AFIK. I believe the latest
gen of SATA drives offer command queuing, once the single greatest advantage
of SCSI. Sure the SCSI interface and protocol is higher bandwidth and more
robust, but with enough spindles, you can close the gap, and add capacity.
One final gripe with Isilion is that it uses Samba. I expected Isilon to
have their own CIFS software stack for this caliber of product. This is not
a deal-breaker, and as good as it is, it cannot touch the proprietary
products offered by NetApp and others. Let¹s face it, while powerful, Samba
is also buggy, but what do you expect, it¹s open soure.
All in all, I would agree with you, ease of management is a good selling
point. I would like to see more powerful reporting tools, however, and more
flexibility with NFS export creation. I hope to test the snapshots feature
this week, and maybe a few other features....
I hear what you say about doing the evals. They are time-consuming and
require a lot of time to plan and implement. If you have large data sets to
move, it is even a bigger pain in the arse. But these tests are necessary
and I cannot understand those that do not bother with them. Next week, after
the holiday, I will start round two, with a Blue Arc eval. As you said, it
did not take much arm-twisting, they are all eager to move product these
days.
I am curious to know if you use Aspera as an alternative to a hardware
WAN-accelerator? I would like to hear more about this as I am looking at
various hardware-based WAN accelerator products at this time for multi-site
replication.
I will keep you posted.
Regards.
Jeff
On 11/19/07 4:05 PM, in article 1645-rush decimal general at seriss decimal com, "Saker
Klippsten" <saker@ZOICSTUDIOS.COM> wrote:
> Hey Jeff
>
> Switch wise:
>
> I don¹t think anything can beat a force10 switch. I don¹t have to think
> twice about recommending them at all.
> Their support is top notch.
> Its got a cisco like syntax and config so if you are famil with cisco this
> should be easy to learn. We have a E1200
> ( Http://www.force10networks.com ) at the heart of our datacenter everything
> plugs into this or the S50¹s via 10GiGE.
>
> We had Extreme Switches and I cant say one good thing about them. Their
> products suck and their Support service sucks.
> The technology used to power them is very very outdated. Its more software
> based and that¹s the primary reason why it locks up
> And cant handle the ³extreme bandwidth² requirements of our field. The
> force10 has not had 1 hiccup since we launched it. They are rock solid.
> The E1200 has a 5TB Back Plane.
>
> Storage wise:
>
> We have over a 100TB¹s of Isilon. Been down the san route and I am not looking
> back.
>
>
> Our Setup..
> Over 1000 Procs and 160 or so workstations.
>
> Isilon has been solid for us. Though like anything its not without its
> hiccups we paid that price 5 years ago when we helped alpha the product but
> today its running great.
>
> -Ease of use and management of the cluster I don¹t think there is a contender
> out there.
> -Snapshots are nice easy to manage.
> -Quota systems, Hard, Soft and Alert.
> -Replication software is fast! ( SyncIQ )
> -Aspera will run native on the Isilons we use this to manage and replicate our
> data to Vancouver disaster recovery cluster in case this log cabin burns down
> :) Its a software wan accelerator ( really amazing ) (It will increase your
> wan link speed 100x )
> -Adding storage is like popping in a removable drive. Connect a few cables and
> power on. Having redundancy is what I like the most.
>
> While I am sure the single Stream performance is not as good as the newer
> bluearcs (though I am sure its right behind it ) You do get redundancy with
> your data but also the systems serving the data. You have the options to
> independently grow your performance over the size of your cluster by adding
> Accelerator nodes or just storage or a combo of both.
>
> -Isilon can monitor your cluster remotely if you opt too and alert you of
> anything they think might fail like a disk that might start to fail they can
> predict this and soft fail ahead of time.
>
> -I know everyone bags on SATA drives but I have had more Fiber drives fail in
> one year on my DDN and Flame Arrays than on our 40 Node Isilon Cluster. Which
> has about 480 drives in it : 4 failed in the last year.
>
> -As you know it uses infiniband as its backend communication protocol and
> infiniband is slowly making its way into many other products we use here on
> the highend compositing side of things. While I do not know for sure it would
> seem likely that isilon could enable front end use of these infiniband ports
> for direct access or plug into another infiniband switch to enable highspeed
> access to the cluster. I know they have 10gige support or will be soon in the
> form of an accelerator node or of the like. :)
> This would put off having to utilize the gige ports on each node and just
> have 2 10gige fiber ports used to feed the network switch.
>
> I am partial to Isilon just how some might be partial to Netapp or Bluearc.
> But I am a sucker for ease of use and management functionality ohh and the
> Blue Lights.
>
> Price wise in the past Isilon has done better against the rest thought I know
> you might be able to get a good deal with Bluearc right now as they are
> competing very aggressively for market share.
>
> What it all comes down to is the environment get an Isilon in and test it out,
> Get a bluearc or a netapp and test it. I know its tuff sometimes to get demos
> in but these days they are all eager to get another sale in. Isilon will put a
> demo in no prob. Bluearc and Netapp it might take a bunch of meetings and
> moving it up the chain of command..
>
> I would love to hear feedback on all storage out there as well. Some people
> don¹t have the time to demo or just don¹t want to deal with it
|