No responses? How about the Coraid folks that monitor this list? Can they
respond?
David
On Wed, Nov 10, 2010 at 2:41 PM, David Leach <tas...@gmail.com> wrote:
> Tracy,
>
> Thank you for responding.
>
> The problem though with "I have never run into any trouble that I am aware
> of" is that that is the problem. Undetected data integrity errors are a risk
> that needs to be quantified and being undetected you won't be aware of it
> until it means the most to you. I would love to see a quantification of what
> the risk level really is. Many network protocols have their own level of
> data integrity built in to protect the data being transferred. TCP has the
> rudimentary checksum which, even though it is weak, will catch some errors
> and does so even today.
>
> My comments about the network interconnect points is that they are points
> to introduce errors as well. I found this paper;
> http://portal.acm.org/citation.cfm?id=347561&dl=ACM&coll=DL&CFID=110304004&CFTOKEN=23505194,
> which is a good read on an analysis done on captured network data over
> several months where the TCP checksum failed but CRC passed.
> They categorized much of the error types and found interesting problems with
> network connection points (routers, switches, etc...) injecting errors on
> the packet streams. CRC passes because it was applied to each link but not
> while it was traversing through a switch or router (and fortunately the
> error was captured by TCP checksum). I even saw an article about using the
> digests on iSCSI for data integrity exactly because the data was traversing
> these network connection points. iSCSI digests are simply a CRC32 being
> applied to the header and/or the payload.
>
> So that has me thinking about AoE and the fact that it doesn't even have an
> optional ability to enable data integrity protection of the ATA data flow.
> Is this because the deployment of solutions using AoE are very tightly
> constrained to a Coraid blade rack connected directly to a large file
> server, no switches or any other network connections in between? Or is it
> because the deployments of AoE are not in enterprise level, mission critical
> environments? The problem potentially gets more acute when you introduce
> jumbo frames.
>
> Don't get me wrong, I love the simplicity of the AoE protocol but I would
> love to understand what companies like Coraid say or present to potential
> customers when they ask these kinds of questions. It also would be real
> simple to add an optional data integrity to AoE but before doing so I would
> still like to understand the characteristics of any potential problems.
>
> David
>
> David
>
> On Wed, Nov 10, 2010 at 1:12 PM, Tracy Reed <tr...@ultraviolet.org> wrote:
>
>> On Wed, Nov 10, 2010 at 11:09:16AM -0600, David Leach spake thusly:
>> > I've been looking at AoE and I'm trying to understand what affect the
>> > Ethernet CRC-32 data integrity checking has on the AoE communications?
>>
>> I too have been wanting to better understand the error correction
>> facilities of
>> AoE. So far I have never run into any trouble that I am aware of.
>>
>> > Has anyone done any analysis or have any response to AoE's ultimate
>> > reliability for missed error detection of bad frames?
>>
>> I suspect it is sufficiently low as to not be an issue. You have to
>> multiply
>> the chance of an error by the chance that the error would not be caught
>> due to
>> being a 32 bit crc.
>>
>> > I think that there is a further problem to understand and that is with
>> > network connection points. AoE is not routable but that doesn't mean you
>> > can't use network switches to interconnect initiators with their AoE
>> targets
>> > and at these switches there seems to be a possible error point
>> introduced
>> > which AoE isn't protecting against? Are there best practices for AoE
>> > installations to protect against these error points?
>>
>> AoE is not layer 3 routable. It is routable in general with spanning tree
>> etc.
>> You can also implement an ethernet tunnel (being careful of MTU concerns
>> etc).
>> I think the genious of AoE vs iSCSI is in adhering to the separation of
>> concerns of each of the layers of the network stack.
>>
>> --
>> Tracy Reed
>> http://tracyreed.org
>>
>
>
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Aoetools-discuss mailing list
Aoetools-discuss@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/aoetools-discuss
