Why is network reading slow?
in [MFC]
|
Prev: Is it a bug in mciSendCommand function?
Next: AFX_MANAGE_STATE() macro disturbs the startup position of the ATL dialog
From: Alexander Grigoriev on 17 Jan 2010 15:42 A large I/O will be split to several requests anyway. The danger of allowing such I/O is that it will require locking enormous amount of memory at once, which may deplete the process' working set and may also deplete the system pools. Several smaller operations issued by the application will avoid that situation, without any measurable loss in throughput. "Joseph M. Newcomer" <newcomer(a)flounder.com> wrote in message news:f1s6l5tkqoo6oiunhp8eo0ts54hdhr5967(a)4ax.com... > If I am running a 32-bit application on Win64, then I cannot write more > than 32 bits worth > of data. But if I am running native 64-bit apps on Win64, there is no > sane reason to > limit me to 4.2GB of data on an I/O operation. I am talking about 64-bit > systems with > native 64-bit apps. I was talking about the silly limitation that makes > it inconvenient > to build even native 64-bit programs and run them! >
From: Joseph M. Newcomer on 17 Jan 2010 17:38 You are confusing an implementation detail with an abstract interface. I am not aware of any requirement that "locking an enormous amount of memory" is necessary; on the contrary, when I teach about how to write device drivers, there is a discussion about the use of "mode Neither", the construction of partial MDLs, the internal ways of breaking up an I/O operation into multiple requests, and making this all invisible to the application programmer. joe On Sun, 17 Jan 2010 12:42:58 -0800, "Alexander Grigoriev" <alegr(a)earthlink.net> wrote: >A large I/O will be split to several requests anyway. The danger of allowing >such I/O is that it will require locking enormous amount of memory at once, >which may deplete the process' working set and may also deplete the system >pools. Several smaller operations issued by the application will avoid that >situation, without any measurable loss in throughput. > >"Joseph M. Newcomer" <newcomer(a)flounder.com> wrote in message >news:f1s6l5tkqoo6oiunhp8eo0ts54hdhr5967(a)4ax.com... >> If I am running a 32-bit application on Win64, then I cannot write more >> than 32 bits worth >> of data. But if I am running native 64-bit apps on Win64, there is no >> sane reason to >> limit me to 4.2GB of data on an I/O operation. I am talking about 64-bit >> systems with >> native 64-bit apps. I was talking about the silly limitation that makes >> it inconvenient >> to build even native 64-bit programs and run them! >> > Joseph M. Newcomer [MVP] email: newcomer(a)flounder.com Web: http://www.flounder.com MVP Tips: http://www.flounder.com/mvp_tips.htm
From: Hector Santos on 17 Jan 2010 19:14 Joseph M. Newcomer wrote: > Illiac-IV was the most important supercomputer of its era, and in > fact its architecture was the prototype for many subsequent > supercomputer architectures. Wow, and I thought I had a decent sense about computer history especially since I had a few years in with Cray machines and mainframes before that. It simply was unheard of by the early 80s - to me, and certainly not something used corporations like Westinghouse back then. -- HLS
From: Joseph M. Newcomer on 17 Jan 2010 19:50 There was exactly one Illiac-IV built. It was always a research machine, and never was intended for production. I checked it out in wikipedia, and the article is essentially what I remember, although I have a number of interesting stories, such as the fact that in those days, the computer center ran 8-to-5 and it was a major effort to get it unlocked so the Illiac-IV programmers could use the equipment. At one point they were simply locked out. Extremely funny stories about how one of the group was a trained locksmith and had immense amounts of fun with the suits. It was intended to have 256 processors, but because it ran vastly over budget, only the 64-processor base configuration was ever built. It was Emitter-Coupled Logic (ECL), a real power hog/heat generator, and the unused sockets had to have dummy plastic "chip cases" installed because gaps created nonlaminar airflow and hotspots. Whenever it ran, it was necessary to first run diagnostics, if they were successful, then run a couple hours of production run, then run diagnostics. If the diagnostics ran successfully the second time, the previous few hours of computation were deemed valid; otherwise, the problem had to be fixed and the computations re-run. joe On Sun, 17 Jan 2010 19:14:08 -0500, Hector Santos <sant9442(a)nospam.gmail.com> wrote: >Joseph M. Newcomer wrote: > > >> Illiac-IV was the most important supercomputer of its era, and in > > > fact its architecture was the prototype for many subsequent > > supercomputer architectures. > >Wow, and I thought I had a decent sense about computer history >especially since I had a few years in with Cray machines and >mainframes before that. It simply was unheard of by the early 80s - to >me, and certainly not something used corporations like Westinghouse >back then. Joseph M. Newcomer [MVP] email: newcomer(a)flounder.com Web: http://www.flounder.com MVP Tips: http://www.flounder.com/mvp_tips.htm
From: Hector Santos on 17 Jan 2010 21:21
What was the preferred language for it? The old story with the Cray, when Seymour was asked by his engineers "What Language doe we use for this new super computer?" Seymour said; "I don't know and care as long as its called FORTRAN!" Basically, Seymour knew that it would be harder to get customers to switch their million dollars investment to the CRAY if people had to use a new language in order to take advantage of its vectorization power. <g> Joseph M. Newcomer wrote: > There was exactly one Illiac-IV built. It was always a research machine, and never was > intended for production. I checked it out in wikipedia, and the article is essentially > what I remember, although I have a number of interesting stories, such as the fact that in > those days, the computer center ran 8-to-5 and it was a major effort to get it unlocked so > the Illiac-IV programmers could use the equipment. At one point they were simply locked > out. Extremely funny stories about how one of the group was a trained locksmith and had > immense amounts of fun with the suits. > > It was intended to have 256 processors, but because it ran vastly over budget, only the > 64-processor base configuration was ever built. It was Emitter-Coupled Logic (ECL), a > real power hog/heat generator, and the unused sockets had to have dummy plastic "chip > cases" installed because gaps created nonlaminar airflow and hotspots. > > Whenever it ran, it was necessary to first run diagnostics, if they were successful, then > run a couple hours of production run, then run diagnostics. If the diagnostics ran > successfully the second time, the previous few hours of computation were deemed valid; > otherwise, the problem had to be fixed and the computations re-run. > joe > > On Sun, 17 Jan 2010 19:14:08 -0500, Hector Santos <sant9442(a)nospam.gmail.com> wrote: > >> Joseph M. Newcomer wrote: >> >> >>> Illiac-IV was the most important supercomputer of its era, and in >>> fact its architecture was the prototype for many subsequent >>> supercomputer architectures. >> Wow, and I thought I had a decent sense about computer history >> especially since I had a few years in with Cray machines and >> mainframes before that. It simply was unheard of by the early 80s - to >> me, and certainly not something used corporations like Westinghouse >> back then. > Joseph M. Newcomer [MVP] > email: newcomer(a)flounder.com > Web: http://www.flounder.com > MVP Tips: http://www.flounder.com/mvp_tips.htm -- HLS |