Re: Memory undectable on Intel server board S5000VCL
Last I used PAE kernels (ages ago) - I saw 1-5% degradation in performance. It was a while ago and I am not even sure I remember the versions any more. What I remember is NOT scientific data. M - "Muli Ben-Yehuda" <[EMAIL PROTECTED]> wrote: > I would like to understand this better, if you or Marc could shed > more > details? In general, I believe you know what they say about "lies, > damn lies, and performance numbers". See some numbers from Ingo > Molnar > which give the PAE overhead at a round 0.0% here: > http://kerneltrap.org/node/2891. YMMV. > >100Hz100.00% >100Hz + PAE: 0.00% > >1000Hz: -1.08% >1000Hz + PAE: -1.08% > > Cheers, > Muli > > = > To unsubscribe, send mail to [EMAIL PROTECTED] with > the word "unsubscribe" in the message body, e.g., run the command > echo unsubscribe | mail [EMAIL PROTECTED] -- ---MAV Marc A. Volovic Swiftouch, LTD [EMAIL PROTECTED] +972-544-676764 = To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
Re: Memory undectable on Intel server board S5000VCL
isnt' there a bigmem kernel for situations just like these? On 8/9/07, Muli Ben-Yehuda <[EMAIL PROTECTED]> wrote: > On Thu, Aug 09, 2007 at 02:59:02PM +0300, Ira Abramov wrote: > > Quoting Muli Ben-Yehuda, from the post of Thu, 09 Aug: > > > > > > Because some PCI devices cannot deal with addresses over 4G. To be > > > able to map the RAM that has physical address over 4G in the CPU's > > > page tables, you need a PAE kernel. As for the penalty on speed, I > > > doubt you'll be able to measure it, and it will be offset by the extra > > > RAM. > > > > According to Marc (who is a man I don't like to disagree with) it's > > a hit of 3-5% on average on grid machines if I recall. > > I would like to understand this better, if you or Marc could shed more > details? In general, I believe you know what they say about "lies, > damn lies, and performance numbers". See some numbers from Ingo Molnar > which give the PAE overhead at a round 0.0% here: > http://kerneltrap.org/node/2891. YMMV. > >100Hz100.00% >100Hz + PAE: 0.00% > >1000Hz: -1.08% >1000Hz + PAE: -1.08% > > Cheers, > Muli > > = > To unsubscribe, send mail to [EMAIL PROTECTED] with > the word "unsubscribe" in the message body, e.g., run the command > echo unsubscribe | mail [EMAIL PROTECTED] > > = To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
Re: Memory undectable on Intel server board S5000VCL
On Thu, Aug 09, 2007 at 02:59:02PM +0300, Ira Abramov wrote: > Quoting Muli Ben-Yehuda, from the post of Thu, 09 Aug: > > > > Because some PCI devices cannot deal with addresses over 4G. To be > > able to map the RAM that has physical address over 4G in the CPU's > > page tables, you need a PAE kernel. As for the penalty on speed, I > > doubt you'll be able to measure it, and it will be offset by the extra > > RAM. > > According to Marc (who is a man I don't like to disagree with) it's > a hit of 3-5% on average on grid machines if I recall. I would like to understand this better, if you or Marc could shed more details? In general, I believe you know what they say about "lies, damn lies, and performance numbers". See some numbers from Ingo Molnar which give the PAE overhead at a round 0.0% here: http://kerneltrap.org/node/2891. YMMV. 100Hz100.00% 100Hz + PAE: 0.00% 1000Hz: -1.08% 1000Hz + PAE: -1.08% Cheers, Muli = To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
Re: Memory undectable on Intel server board S5000VCL
Hi Ira, I bumped into the same question some time ago - sending to iglu.org.il silently failed. Here's why: http://www.mail-archive.com/linux-il@cs.huji.ac.il/msg48796.html Boaz. Ira Abramov wrote: > annoying! I just figured out why my posts to the list have disappeared > in the last couple of months... the @iglu.org.il alias no longer points > to huji. why is that? Uff... NM, here is the question again... > > - Forwarded message from Ira Abramov > > To: IGLU <[EMAIL PROTECTED]> > Subject: Memory undectable on Intel server board S5000VCL > Date: Mon, 11 Jun 2007 12:30:22 +0300 > > Howdie folks. > > I just transplanted my server to a new 1U Pizza. An amazing bargain. A > Quad-core Xeon on a dual-CPU Intel board (room for enhancement!) with 4G > of FBDIMMs, and two 500GB disks, all came to just over 10K NIS after > taxes. > > The problems I see so far: hardware raid doesn't work in Debian (There > are binary-only drivers for RHEL and SuSE EL), there are special > editions of the e1000 drivers that are supposed to speed up the network > through some partial off loading, I haven't read what they do yet. but > the problem I am having is the PCI memory mapping. It's set at "2.5GB" by > default in the BIOS, and Debian Etch comes up and indeed sees only > 2.5GB. I have set it up to 3.5G to lose less RAM, but I would still like > to use the rest of the RAM! I tried booting with ram=4G but still the > kernel sees only 3.5. I guess I need to use memmap= as well, but I don't > have that information. > > the only clue is this tiny line in the BIOS docs: > > -PAE mode and the equivalent Linux mode must be set to see > 2.5GB > of memory. > > but running the debian "bigmem" kernel with 4G or less has an unwanted > penalty on speed, doesn't it? > > any advice? > > = To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
Re: Memory undectable on Intel server board S5000VCL
Quoting Muli Ben-Yehuda, from the post of Thu, 09 Aug: > > Because some PCI devices cannot deal with addresses over 4G. To be > able to map the RAM that has physical address over 4G in the CPU's > page tables, you need a PAE kernel. As for the penalty on speed, I > doubt you'll be able to measure it, and it will be offset by the extra > RAM. According to Marc (who is a man I don't like to disagree with) it's a hit of 3-5% on average on grid machines if I recall. welp... we'll wait for the need for that extra 1/2 a gig to arrive and then reconsider using it. for now it's doing OK :) -- Late bloomer Ira Abramov http://ira.abramov.org/email/ = To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
Re: Memory undectable on Intel server board S5000VCL
Huh? your mail has been received well (this and the first one you sent) Thanks, Hetz On 09/08/07, Ira Abramov <[EMAIL PROTECTED]> wrote: > annoying! I just figured out why my posts to the list have disappeared > in the last couple of months... the @iglu.org.il alias no longer points > to huji. why is that? Uff... NM, here is the question again... > > - Forwarded message from Ira Abramov > > To: IGLU <[EMAIL PROTECTED]> > Subject: Memory undectable on Intel server board S5000VCL > Date: Mon, 11 Jun 2007 12:30:22 +0300 > > Howdie folks. > > I just transplanted my server to a new 1U Pizza. An amazing bargain. A > Quad-core Xeon on a dual-CPU Intel board (room for enhancement!) with 4G > of FBDIMMs, and two 500GB disks, all came to just over 10K NIS after > taxes. > > The problems I see so far: hardware raid doesn't work in Debian (There > are binary-only drivers for RHEL and SuSE EL), there are special > editions of the e1000 drivers that are supposed to speed up the network > through some partial off loading, I haven't read what they do yet. but > the problem I am having is the PCI memory mapping. It's set at "2.5GB" by > default in the BIOS, and Debian Etch comes up and indeed sees only > 2.5GB. I have set it up to 3.5G to lose less RAM, but I would still like > to use the rest of the RAM! I tried booting with ram=4G but still the > kernel sees only 3.5. I guess I need to use memmap= as well, but I don't > have that information. > > the only clue is this tiny line in the BIOS docs: > > -PAE mode and the equivalent Linux mode must be set to see > 2.5GB > of memory. > > but running the debian "bigmem" kernel with 4G or less has an unwanted > penalty on speed, doesn't it? > > any advice? > > -- > Another brick in the wall > Ira Abramov > http://ira.abramov.org/email/ > > = > To unsubscribe, send mail to [EMAIL PROTECTED] with > the word "unsubscribe" in the message body, e.g., run the command > echo unsubscribe | mail [EMAIL PROTECTED] > > -- Skepticism is the lazy person's default position. my blog (hebrew): http://benhamo.org = To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
Re: Memory undectable on Intel server board S5000VCL
On Thu, Aug 09, 2007 at 11:23:09AM +0300, Ira Abramov wrote: > the problem I am having is the PCI memory mapping. It's set at > "2.5GB" by default in the BIOS, and Debian Etch comes up and indeed > sees only 2.5GB. I have set it up to 3.5G to lose less RAM, but I > would still like to use the rest of the RAM! I tried booting with > ram=4G but still the kernel sees only 3.5. I guess I need to use > memmap= as well, but I don't have that information. > > the only clue is this tiny line in the BIOS docs: > > -PAE mode and the equivalent Linux mode must be set to see > > 2.5GB of memory. > > but running the debian "bigmem" kernel with 4G or less has an > unwanted penalty on speed, doesn't it? For all intents and purposes, you have more than 4G of RAM. More accurately, some of your RAM is only addressable using addresses that are over 4G. The reason why is that your physical memory addresses look like this (simplified): 0<---RAM--->3.5G<---PCI HOLE---><---more RAM over 4G---> Because some PCI devices cannot deal with addresses over 4G. To be able to map the RAM that has physical address over 4G in the CPU's page tables, you need a PAE kernel. As for the penalty on speed, I doubt you'll be able to measure it, and it will be offset by the extra RAM. Cheers, Muli = To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
