|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] RE: iscsi : DataPDULength can differ in each direction.Ayman, A few people have raised the issue of added complexity because of the CRC calculation having to be done on different PDU sizes and I don't understand the difficulty. No matter what maximum PDU size one supports one has to do the CRC on the actual PDUs which are sent which will vary. One doesn't always send the maximum - sometimes the write or read is for less data than that. Anyway the calculation is done one byte/word/(other chunk size) at a time until one gets to the end of the PDU regardles of the length. I would appreciate if someone could clarify why different max PDU sizes complicate CRC calculation. Pat -----Original Message----- From: Ayman Ghanem [mailto:aghanem@cisco.com] Sent: Wednesday, October 03, 2001 11:38 AM To: ips@ece.cmu.edu Subject: RE: iscsi : DataPDULength can differ in each direction. I think this will add unnecessary complexity, specially for data CRCs having to be calculated based on two different PDU sizes for Reads and Writes. Also, one end may choose to do data digests in software. If the other end decides to use a really large PDU length (and the first doesn't have a say about it), this could be a problem. -Ayman > -----Original Message----- > From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of > Santosh Rao > Sent: Wednesday, October 03, 2001 12:32 PM > To: Dev - Platys; Julian Satran > Cc: Sandeep Joshi; ips@ece.cmu.edu > Subject: iscsi : DataPDULength can differ in each direction. > > > Deva, Julian & All, > > I think we have a more fundamental problem here, that spans beyond the > issue of which negotiation model should be used for numerical & binary > key negotiation. This problem needs to be addressed seperately. > > The DataPDULength can and should be allowed to be different in each > direction. I -> T direction PDUs should be allowed to use a different > PDU Length than T -> I direction PDUs. > > Each side should be allowed to specify the DataPDULength it will be > using and there should be no attempt to negotiate this value. Rather, > the key is exchanged in either direction to inform the other side as to > what sized PDUs it should expect. > > This is somwhat akin to a protocol's MTU definition or a max_frame_size > definition (ex : FC PLOGI Receive Data Field Size in its common service > parameters.). THese values are allowed to be different going in each > direction, allowing for implementation specific quirks that restrict the > use of certain values. (ex : some DMA bugs may require implementations > to have their DataPDULength to be a cache line size shorter, etc.) > > I suggest the definition of the DataPDULength key be re-phrased taking > the above into account. > > Regards, > Santosh > > Dev - Platys wrote: > > > > Santosh, > > > > The point is, if the target is not capable of supporting 16K > (it is just an > > example that is being discussed), then it indicates that is > what it wants. > > Initiator can close the connection. (No reject is necessary though). > > > > However, if the target knows it can support the initiator > negotiated value, > > it could return it. > > > > Again, Yes, implicitly it becomes the responder selects logic. > > > > Deva > > Adaptec > > > > -----Original Message----- > > From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of > > Sandeep Joshi > > Sent: Wednesday, October 03, 2001 7:29 AM > > To: ips@ece.cmu.edu > > Subject: Re: iscsi : numerical negotiation wording is ambiguous > > > > The first & third example below seem illogical. > > > > The responder should not be sending 24K if the initial > > sender has chosen a value of 16K, since there is no > > possibility of that 24K being accepted (unless we want > > three way exchanges for every negotiation??) > > > > The problem of a Reject message does not arise since the > > responder should only be choosing something less than 16K. > > So I guess this puts me in the "responder selects logic" > > camp. > > > > -Sandeep > > > > Julian Satran wrote: > > > > > > An example: > > > 08.txt logic > > > i->t DataPDULength=16k > > > t->i DataPDULength= 24k > > > > > > Selected value is 16k > > > > > > i->t DataPDULength=16k > > > t->i DataPDULength= 8k > > > > > > Selected value is 8k > > > > > > t->i DataPDULength=16k > > > i->t DataPDULength= 24k > > > > > > Selected value is 16k > > > > > > t->i DataPDULength=16k > > > i->t DataPDULength= 8k > > > > > > Selected value is 8k > > > > > > ----- > > > > > > Responder selects logic > > > > > > i->t DataPDULength=16k > > > t->i DataPDULength= 24k > > > > > > Error - Initiator Reject - Closes connection > > > > > > i->t DataPDULength=16k > > > t->i DataPDULength= 8k > > > > > > Selected value is 8k > > > > > > t->i DataPDULength=16k > > > i->t DataPDULength= 24k > > > > > > Error Target has to terminate with parameter error > > > > > > t->i DataPDULength=16k > > > i->t DataPDULength= 8k > > > > > > Selected value is 8k > > > > > > "Dev - Platys" > > > <deva@platys.com> To: "Julian Satran" > > > <Julian_Satran@il.ibm.com>, > > > 01-10-01 23:46 <ips@ece.cmu.edu> > > > Please respond to "Dev - cc: > > > Platys" Subject: RE: iscsi : > > > numerical negotiation wording is ambiguous > > > > > > > > > > > > Julian, > > > > > > I am not sure why a 'REJECT' is required. > > > Can you please explain why is this required? > > > > > > I am with Santosh in this. > > > > > > >There is NO need for any REJECT in the above >case. If the initiator > > > >is'nt satisfied with the value returned by the >originator and cannot > > > >function with the negotiated values, it can simply >close the TCP > > > >connection. There is no need to send any >REJECT. > > > > > > Thanks > > > > > > Deva > > > Adaptec > > > > > > -----Original Message----- > > > From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of > > > Julian Satran > > > Sent: Monday, October 01, 2001 1:28 PM > > > To: ips@ece.cmu.edu > > > Subject: Re: iscsi : numerical negotiation wording is ambiguous > > > > > > Santosh, > > > > > > We are getting nowhere. Even if requester is doing it's stuff - the > > > requester will have to check and be prepared for a mistake. The coding > > > will require a reject. > > > > > > Julo > > > > > > Santosh Rao > > > <santoshr@cup.hp.com> To: ips@ece.cmu.edu > > > Sent by: cc: "Sanjeev Bhagat > > > santoshr@cup.hp.com (TRIPACE/Zoetermeer)" <sbhagat@tripace.com>, > > > Julian Satran/Haifa/IBM@IBMIL > > > 01-10-01 20:37 Subject: Re: iscsi : numerical > > > Please respond to negotiation wording is ambiguous > > > Santosh Rao > > > > > > > > > Julian & Sanjeev, > > > > > > Responding to both your mails...... > > > > > > Julian : > > > > > > I think you may have mis-interpreted my comments. I believe Sanjeev > > > has > > > clarified the intent of my suggestions. > > > > > > I am *not* suggesting that the responder send back its values and > > > these > > > be blindly imposed on the originator. On the contrary, my suggestion > > > is > > > that the computation of the result of the negotiation (higher or lower > > > of the 2 values) be only performed by the responder and sent back to > > > the > > > originator. > > > > > > The result of the negotiation is the same in both cases and there is > > > no > > > REJECT required in my case nor yours. The difference is the advantages > > > I've stated in my model. > > > > > > Sanjeev, in response to your comment : > > > > > > " [Sanjeev Bhagat (TRIPACE/Zoetermeer)] Although there is > > > > no reject , but it can be a problem in future . > > > > Consider your example of DataPDULength in your own > > > > message. Suppose offering party sends a value of 16,384 > > > > (this is also lowest value it can send) and responding > > > > party responds with 8,192. In this case the offering > > > > party will have to reject the negotiation. So a reject > > > > message is needed in this case also. " > > > > > > There is NO need for any REJECT in the above case. If the initiator > > > is'nt satisfied with the value returned by the originator and cannot > > > function with the negotiated values, it can simply close the TCP > > > connection. There is no need to send any REJECT. > > > > > > Thanks, > > > Santosh > > > > > > > "Sanjeev Bhagat (TRIPACE/Zoetermeer)" wrote: > > > > > > > > Thanks Julian, please find my further comments in the message > > > > > > > > -----Original Message----- > > > > From: Julian Satran [mailto:Julian_Satran@il.ibm.com] > > > > Sent: Sunday, September 30, 2001 6:07 PM > > > > To: ips@ece.cmu.edu > > > > Subject: RE: iscsi : numerical negotiation wording is > > > > ambiguous > > > > > > > > Sanjeev, > > > > > > > > I am not sure clear I made the tiny diference between what I > > > > say and what Santosh said: > > > > > > > > Santosh says: > > > > > > > > 1. a requester sends A=valuex > > > > 2. a responder sends B=valuey > > > > 3. the responder assumes that the value y is the correct > > > > value and so does an external observer > > > > [Sanjeev Bhagat (TRIPACE/Zoetermeer)] I would rather > > > > say it this way the responder computes the value with > > > > its own supported values and responds with a value y > > > > which the responder thinks is correct and so does an > > > > external observer > > > > 4. the requester checks that valuey is conformant (he will > > > > not believe it) and will reject it if not conformant > > > > else it will accept it > > > > [Sanjeev Bhagat (TRIPACE/Zoetermeer)] Although correct, > > > > but it is highly unlikely for the responder to reject > > > > the final result because the rule states that (lower or > > > > higher of two will be the result) and so the offering > > > > party should be able to accept the lower or higher > > > > range as defined by rule. In case the key is dependent > > > > upon some range of fixed values then the negotiation > > > > should be performed as list negotiation and not > > > > numerical negotiation. > > > > > > > > This might be what you "conventionally" do - I don't > > > > and that shows that convention like morals are a matter > > > > of geography :-) > > > > [Sanjeev Bhagat (TRIPACE/Zoetermeer)] :-) > > > > > > > > The advantage of this set of rules is that it allows an > > > > external observer to know what was selected without > > > > knowing the rules > > > > [Sanjeev Bhagat (TRIPACE/Zoetermeer)] Even in this > > > > case, I guess that the external observer has to know > > > > the rules to double check the value is correct or not > > > > The disadvantage is that messages have to be "built", > > > > an additional reject states exists and MOST IMPORTANT > > > > you need both messages > > > > > > > > In what I said: > > > > > > > > 1. The requester sends A=valuex > > > > 2. The Responder has to send either nothing (if valuex > > > > is enough on both sides to compute the result like in > > > > the case in which the function is a Boolean AND and the > > > > value is NO) or valuey > > > > 3. Both the requestor and responder have to compute the > > > > value (they have to know the rules anyhow) and so does > > > > the external observer > > > > > > > > The disadvantage is that the external observer has to > > > > know the rules > > > > The advantage is that there is no reject, in binary > > > > negotiations you can go away with shorter negotiations > > > > and you can set strings at fixed values. > > > > [Sanjeev Bhagat (TRIPACE/Zoetermeer)] Although there is > > > > no reject , but it can be a problem in future . > > > > Consider your example of DataPDULength in your own > > > > message. Suppose offering party sends a value of 16,384 > > > > (this is also lowest value it can send) and responding > > > > party responds with 8,192. In this case the offering > > > > party will have to reject the negotiation. So a reject > > > > message is needed in this case also. > > > > > > > > > > > > Sanjeev > > > > "Sanjeev Bhagat > > > > (TRIPACE/Zoetermeer)" To: "'Santosh > > > Rao'" > > > > <sbhagat@tripace.com> <santoshr@cup.hp.com>, Julian > > > > Sent by: owner-ips@ece.cmu.edu Satran/Haifa/IBM@IBMIL > > > > cc: > > > ips@ece.cmu.edu > > > > 30-09-01 16:32 Subject: RE: > > > iscsi : > > > > Please respond to "Sanjeev numerical negotiation wording > > > is > > > > Bhagat (TRIPACE/Zoetermeer)" ambiguous > > > > > > > > > > > > > > > > All, > > > > > > > > I agree with Santosh that the responding party must respond > > > > the result of > > > > the negotiation as compared with the value from offering > > > > party. All > > > > negotiations in SCSI like (sync, disconnect etc) are also > > > > done the same way. > > > > I also object to Julian's reason of a simple minded target > > > > which wants to > > > > send certain fixed values only. In case a simple minded > > > > target identifies a > > > > value which it cannot support or is less than the value a > > > > target can > > > > support, then there should be a method for target to reject > > > > such a > > > > negotiation and the default values be accepted on both side > > > > as a result of > > > > negotiation. > > > > > > > > 1 Because even if simple minded target sends its fixed value > > > > which is > > > > greater than the value sent by offering party then the final > > > > result of > > > > negotiation will be taken as ( lesser of the two) and in > > > > this case target > > > > which which cannot support the lower value will produce some > > > > illegal > > > > results. > > > > > > > > 2. if simple minded target wants to send its own value and > > > > wants it to be > > > > accpeted then the responding party is not negotiating but > > > > forcing the result > > > > on initiator(this method should not be allowed unless > > > > explicitly mentioned). > > > > > > > > however if there is another proposal of numeric negotiation > > > > in which the > > > > responding party can force its result to be over ridden by > > > > the offering > > > > party's result then it is acceptable for offering party and > > > > responding party > > > > to send there own supported key-value result and it can then > > > > be computed at > > > > offering party's end. > > > > > > > > IMP: (May be I am missing something here) I do not see how a > > > > numeric > > > > negotiation can be rejected. Is it possible to reject such > > > > kind of > > > > negotiaion? > > > > > > > > Sanjeev > > > > > > > > -----Original Message----- > > > > From: Santosh Rao [mailto:santoshr@cup.hp.com] > > > > Sent: Friday, September 28, 2001 11:15 PM > > > > To: Julian Satran > > > > Cc: ips@ece.cmu.edu > > > > Subject: Re: iscsi : numerical negotiation wording is > > > > ambiguous > > > > > > > > Julian & All, > > > > > > > > I request the group to take a close look at this negotiation > > > > process > > > > again and [re-]evaluate the alternative being proposed. > > > > > > > > Further, it appears that the login stage negotiation is > > > > also following > > > > the same algorithm as the login key negotiation, wherein > > > > originator & > > > > responder offer their respective values and both sides need > > > > to determine > > > > the result of the negotiation. i.e. both initiator and > > > > target need to > > > > compare their NSG with the other party's NSG and pick the > > > > lower of the > > > > 2. > > > > > > > > I suggest that both the login key negotiation and the login > > > > stage > > > > negotiation follow the policy that the originator offers a > > > > value and the > > > > responder picks the result of the negotiation based on (the > > > > offered > > > > value & its own value). The value picked by the responder is > > > > sent back > > > > to the originator. > > > > > > > > This model has the following advantages : > > > > > > > > 1) Only one side picks the result of the negotiaton. Hence, > > > > the 2 sides > > > > cannot go out of sync on the value picked. > > > > > > > > 2) The originator knows the negotiated result at the the > > > > responder since > > > > the responder sends back the result of the negotiation. > > > > > > > > 3) This model is easier to debug because of (1) & (2). > > > > > > > > 4) Less code since only 1 party (responder) needs to perform > > > > the > > > > computation to pick the result of the negotiation. > > > > > > > > 5) This scheme leaves less room for interop problems and > > > > mis-interpretation since it is the more familiar negotiation > > > > technique > > > > that is in use in most other mass storage protocols. (ex : > > > > FC PLOGI, FC > > > > PRLI, etc). From a first reading of the current negotiation > > > > scheme, it > > > > is'nt readily apparent that the currently defined model is > > > > different > > > > from the above and requires both sides to be picking the > > > > result of the > > > > negotiation, instead of just the responder. > > > > > > > > Comments ? > > > > > > > > Thanks, > > > > Santosh > > > > > > > > > > -- > > > ################################## > > > Santosh Rao > > > Software Design Engineer, > > > HP-UX iSCSI Driver Team, > > > Hewlett Packard, Cupertino. > > > email : santoshr@cup.hp.com > > > Phone : 408-447-3751 > > > ################################## > > -- > ################################## > Santosh Rao > Software Design Engineer, > HP-UX iSCSI Driver Team, > Hewlett Packard, Cupertino. > email : santoshr@cup.hp.com > Phone : 408-447-3751 > ################################## >
Home Last updated: Fri Oct 05 12:17:48 2001 7070 messages in chronological order |