Okay, thanks, that makes sense... Arjan wrote said that he doubted this was my problem, and that the problem does still exist, but will be fixed in 2.4.20ish. Now, it seems odd to me that the kernel should swap anything to disk on account of the buffers, but not I don't know much about the intricacies of the kernel. I too have read that disabling swap is a bad thing, but no one has explained why not to me satisfactorily. Do you know? (sorry that this is somewhat offtopic). Checking right now, I've got a whipping 200k of the swap files in use. In any case, I had 256MB of RAM. I had another 256mb of pc133 lying around which I wasn't using, becuase it's cl3, and the stuff I've got in my system is cl2. I changed the latency to 3 and put all the RAM in. Since then, I haven't been able to make the system freeze. But, on one copy attempt, I got a segmentation fault. Which strikes me as very, very odd. In any case, I'm now going to pull the old RAM out, see what happens with 256MB and the swap on the SCSI disk... ----- Original Message ----- From: "Bas Hulsken" <bhulsken@xxxxxxxxxxxxxx> To: <ataraid-list@xxxxxxxxxx> Sent: Monday, August 19, 2002 2:13 PM Subject: Re: HPT370 RAID0 hangs under heavy load > Hi Berck, > > No I don't mean cache, sorry I was a bit drunk&very tired when I wrote > that:) What I meant was swap.. the problem I think you have is that if > your memory is full, you have to swap pages of memory to disk, to get > more free memory. But for swapping pages to disk you need to write to > disk. I think you have your swap mounted on your ataraid partition > right? Now if the (current) ataraid driver has to write to disk, it > needs to allocate memory for buffers. But the memory was full, thats why > the thing started swapping pages to the disk in the first place. So > memory allocation fails, & no more memory can be freed by swapping to > disk because more memory needs to be allocated to swap to disk. What we > have here is a deadlock, the ataraid driver is waiting for the > virtualmemory system to get some free memory and the virtualmemory > system is waiting for the ataraid driver to write some to be swapped out > pages to disk. I hope I am a bit more clear now:). > > However this was a problem half a year ago (when I had the same problem > you describe) so I'm not completely sure if it is fixed or not. Arjen de > Ven probably knows, he's the maintainer of the ataraid driver (if I > remember correctly) he's a busy man so be patient, but he reads this > mailing list so I hope he can confirm/deny the story I just told > you.(Arjen?). BTW, as to the solutions I proposed, buying more memory > just avoids swapping, and thus the deadlock (when I bought the 1.5GB it > was very cheap, prices are high again these days). Putting swap on a > disk with another driver (just a disk to another ide/scsi controller) > might help too. Disabling swap will probably work to, but it works only > if you have a lot of memory, and the kernel guys say NEVER disable swap, > so nothings guaranteed. > > Hope I helped you a bit more this time:) > > regards > Bas Hulsken > > > On Mon, 2002-08-19 at 18:02, ataraid-list-request@xxxxxxxxxx wrote: > > > Message: 2 > > From: "Berck E. Nash" <berck@xxxxxxxxxxxx> > > To: <ataraid-list@xxxxxxxxxx> > > Subject: Re: Ataraid-list digest, Vol 1 #285 - 12 msgs > > Date: Sun, 18 Aug 2002 17:29:08 -0400 > > Reply-To: ataraid-list@xxxxxxxxxx > > > > Thanks! I don't really understand the description, entirely, but I'm so glad > > to see a solution to my problem listed. It seems that disabling disk > > caching would slow things down quite a lot? Is there way to do this easily > > so that I can verify that this is, indeed the problem? > > > > 1.5GB of ram, seems a somewhat pricey solution to the problem... Shall I > > email this van de Ven character and see what he has to say? > > > > > _______________________________________________ > > Ataraid-list@xxxxxxxxxx > https://listman.redhat.com/mailman/listinfo/ataraid-list >
