[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: RE: [kmip] Groups - Sun 64-bit Binary Alignment Proposal v2 (Sun 64-bit Binary Alignment Proposal.pdf) uploaded
Matt, When I look at your two most recent proposals posted on the 6th and the 1st, here is a summary of the different values in section 9.1.1: Proposal Item Tag Item Type 32 Bit 32 bits 32 bits 64 bit 3 byte 1 byte Please help me understand where the 16 bit fields are. Do you have any data as to how long it will take to process 32 bit or 64 bit fields with 64 bit processors? Won't the difference be in the microsecond range? I don't think they will amount to much difference. Thanks, Scott -----Original Message----- From: Matthew.Ball@Sun.COM [mailto:Matthew.Ball@Sun.COM] Sent: Wednesday, May 06, 2009 11:37 AM To: Scott Kipp Cc: kmip@lists.oasis-open.org Subject: Re: [kmip] Groups - Sun 64-bit Binary Alignment Proposal v2 (Sun 64-bit Binary Alignment Proposal.pdf) uploaded Hi Scott, See responses below: Scott Kipp wrote: > Matt, > > This 64-bit proposal only has a 1-octet type field while the 32-bit > proposals type field has 4 octets. I would have thought the 64-bit > proposal would expand the type field to a 4 octet field to be in > alignment with the 32-bit proposal. I think the 64-bit proposal will > run into scalability problems if the type field only has 256 values as > others have said. > [Matt] I think you're confusing the Type for the Tag field here. Previous, we were concerned with limiting the _Tag_ field to 8 usable bits. I don't know of anyone worried about expanding the Type field beyond 8 bits because Types are fundamental to the parser and are unlikely to change (at least without breaking all backwards compatibility). Conversely, we expect moderate expansion of the Tag field, and both the 32-bit and 64-bit proposals allow 16-bits for this (i.e., both the 32-bit and 64-bit proposals are identical in this regard). > All, > > I won't be on the call tomorrow, so I want to say why I voted for the > 32-bit alignment. There are many ways to specify this and our goal > should be to pick the optimal one. The two options we are considering > is to optimize the bits on the wire (and thus the bits in storage) > with the 32-bit alignment or optimize the processor performance for > 64-bit processors with 64-bit alignment. > [Matt] I want to be very clear that the 64-bit alignment proposal does not make it hard for 32-bit processors. The 64-bit alignment proposal is equally suitable to either 32-bit or 64-bit processors. Let me clarify, if possible: In the 64-bit proposal, the Tag and Type fields are combined into a single 32-bit word, making it optimal for 32-bit processors. The length field is also 32-bits, making it optimal for 32-bit processors. The combination of the Tag, Type, and Length fields fits nicely into 64-bits, making it additionally optimal for 64-bit processors. In the 64-bit proposal, the data field is padded out to multiples of 8 bytes, which still works find on a 32-bit processor. Thanks, -Matt
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]