Skip to contentSkip to conversation

2.4.23 includes Andrea's VM?

DISCUSSION Ian Soboroff 5 participants 10 messages Dec 3, 2003 – Dec 6, 2003

lore
Share & tools
lore
#0 lore raw

I have a machine with 12GB of RAM, and I've been running a 2.4.22-era kernel with Andrea's patches on it, otherwise it dies from lack of lowmem.

The latest -aa patch is for 2.4.23-pre6, but I see in the 2.4.23 Changelog that at least some bits of Andrea's VM were merged. Should I be able to run a vanilla 2.4.23 on this box?

(Well, vanilla modulo XFS, but I know how to apply patches ;-)

For the record, I tried hand-merging the last -aa patch into 2.4.22-rc5 last week, but it wasn't 100% stable so I may have fudged something. (It would lock up when the console blanked...)

Ian

#1 lore raw
Quoted 9 lines · Ian Soboroff

I have a machine with 12GB of RAM, and I've been running a 2.4.22-era kernel with Andrea's patches on it, otherwise it dies from lack of lowmem.

The latest -aa patch is for 2.4.23-pre6, but I see in the 2.4.23 Changelog that at least some bits of Andrea's VM were merged. Should I be able to run a vanilla 2.4.23 on this box?

A good amount of the VM was merged into 2.4.23-pre3, so the -aa patches against pre6 should show you what is missing.

That said, I have seen a report that no stock 2.4 kernel would run well > 4GB memory until 2.4.23, but he didn't say how much memory he had.

#2 lore raw
Quoted 11 lines · On Wed, 3 Dec 2003 09:44:45 -0800 Mike Fedyk <[email protected]> wrote:
Quoted 9 lines · Ian Soboroff

I have a machine with 12GB of RAM, and I've been running a 2.4.22-era kernel with Andrea's patches on it, otherwise it dies from lack of lowmem.

The latest -aa patch is for 2.4.23-pre6, but I see in the 2.4.23 Changelog that at least some bits of Andrea's VM were merged. Should I be able to run a vanilla 2.4.23 on this box?

Quoted 5 lines

A good amount of the VM was merged into 2.4.23-pre3, so the -aa patches against pre6 should show you what is missing.

That said, I have seen a report that no stock 2.4 kernel would run well > 4GB memory until 2.4.23, but he didn't say how much memory he had.

6 GB

Lowmem looks pretty free though. This lets me suspect 12 GB should be working out, too. If it fails in your tests anyways, please let me know ...

Regards,
Stephan

#3 lore raw
In reply to Ian Soboroff
Quoted 8 lines · Ian Soboroff

I have a machine with 12GB of RAM, and I've been running a 2.4.22-era kernel with Andrea's patches on it, otherwise it dies from lack of lowmem.

The latest -aa patch is for 2.4.23-pre6, but I see in the 2.4.23 Changelog that at least some bits of Andrea's VM were merged. Should I be able to run a vanilla 2.4.23 on this box?

It's probably going to work an order of magnitude better thanks especially to the lower_zone_reserve algorithm.

However I'd still recommend to use my tree, the last two critical bits you need from my tree are inode-highmem and related_bhs. Those two are still missing, and you probably need them with 12G.

I'm going to release a 2.4.23aa1 btw, that will be the last 2.4-aa.

#4 lore raw

2.6-aa? was: 2.4.23 includes Andrea's VM?

Quoted 1 line · Andrea Arcangeli

I'm going to release a 2.4.23aa1 btw, that will be the last 2.4-aa.

So are you going to have a 2.6-aa now? :)

#5 lore raw
In reply to Andrea Arcangeli
Quoted 8 lines · Andrea Arcangeli

It's probably going to work an order of magnitude better thanks especially to the lower_zone_reserve algorithm.

However I'd still recommend to use my tree, the last two critical bits you need from my tree are inode-highmem and related_bhs. Those two are still missing, and you probably need them with 12G.

I'm going to release a 2.4.23aa1 btw, that will be the last 2.4-aa.

I found 10_inode-highmem-2 in the 2.4.23pre6aa3 directory, but I couldn't find any related_bhs one. Am I looking in the wrong place?

I'd wait for -aa1, but I want to try the updated aic7xxx driver in 2.4.23 sooner rather than later.

Ian

#6 lore raw
Quoted 5 lines · Ian Soboroff
Quoted 3 lines · Andrea Arcangeli

It's probably going to work an order of magnitude better thanks especially to the lower_zone_reserve algorithm.

Quoted 8 lines
Quoted 5 lines

However I'd still recommend to use my tree, the last two critical bits you need from my tree are inode-highmem and related_bhs. Those two are still missing, and you probably need them with 12G.

I'm going to release a 2.4.23aa1 btw, that will be the last 2.4-aa.

I found 10_inode-highmem-2 in the 2.4.23pre6aa3 directory, but I couldn't find any related_bhs one. Am I looking in the wrong place?

the latter is not a self contained patch unfortunately (it could be in theory but it isn't in practice), and there's no way I can invest effort in 2.4 to extract it now (infact I need to check that 2.6 addresses that instead). But you can find it in the 05_vm_26-rest-1 patch. Ideally if you apply all the 05_vm_* and the inode-highmem (solving possible rejects, or applying only the patch with dependencies), you should be fine then.

Quoted 2 lines

I'd wait for -aa1, but I want to try the updated aic7xxx driver in 2.4.23 sooner rather than later.

Just to try the driver you can go with plain 2.4.23 right now, the inode and bh troubles showup in a few days normally, not in a few minutes, however it depends on your workload. For example with 12G you probably want to avoid updatedb until you apply all the vm fixes.

#7 lore raw
Quoted 4 lines · Andrea Arcangeli

Just to try the driver you can go with plain 2.4.23 right now, the inode and bh troubles showup in a few days normally, not in a few minutes, however it depends on your workload. For example with 12G you probably want to avoid updatedb until you apply all the vm fixes.

My workload eats inodes and bhs for lunch. I'll try merging 05_vm_*. Thanks!

Ian

#8 lore raw
In reply to Andrea Arcangeli

In article <[email protected]>, Andrea Arcangeli <[email protected]> wrote:

| On Wed, Dec 03, 2003 at 09:51:36AM -0500, Ian Soboroff wrote:
| > 
| > I have a machine with 12GB of RAM, and I've been running a 2.4.22-era
| > kernel with Andrea's patches on it, otherwise it dies from lack of
| > lowmem.  
| > 
| > The latest -aa patch is for 2.4.23-pre6, but I see in the 2.4.23
| > Changelog that at least some bits of Andrea's VM were merged.  Should
| > I be able to run a vanilla 2.4.23 on this box?
| 
| It's probably going to work an order of magnitude better thanks
| especially to the lower_zone_reserve algorithm.
| 
| However I'd still recommend to use my tree, the last two critical bits
| you need from my tree are inode-highmem and related_bhs. Those two are
| still missing, and you probably need them with 12G.
| 
| I'm going to release a 2.4.23aa1 btw, that will be the last 2.4-aa.

I'd like to thank you for all the great VM work you've done, and all the help you gave me off-list back when 1GB was still "a hell of a lot of RAM" and responded to tuning bdflush.

Since 2.4 is now frozen it's probably the last -aa release ever needed, in any case. Future releases, if any, are unlikely to change anything which would keep the patches from working.

Signature · 3 lines
bill davidsen <[email protected]>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.
#9 lore raw
In reply to Andrea Arcangeli
Quoted 10 lines · On Wed, 3 Dec 2003 19:37:19 +0100 Andrea Arcangeli <[email protected]> wrote:
Quoted 8 lines · Ian Soboroff

I have a machine with 12GB of RAM, and I've been running a 2.4.22-era kernel with Andrea's patches on it, otherwise it dies from lack of lowmem.

The latest -aa patch is for 2.4.23-pre6, but I see in the 2.4.23 Changelog that at least some bits of Andrea's VM were merged. Should I be able to run a vanilla 2.4.23 on this box?

Quoted 8 lines

It's probably going to work an order of magnitude better thanks especially to the lower_zone_reserve algorithm.

However I'd still recommend to use my tree, the last two critical bits you need from my tree are inode-highmem and related_bhs. Those two are still missing, and you probably need them with 12G.

I'm going to release a 2.4.23aa1 btw, that will be the last 2.4-aa.

Let us try these please and have them (at last) included in mainline. Thanks for your continous work.

Regards,
Stephan