ircmaxell's questions

Let me start off by explaining why I want to do this. Everything was running fine. I imported a snapshot of a MySQL DB on another server in prep for setting up master-master replication (this one will become the primary in the array once it's setup). I had turned MySQL slave replication and it was catching up. I also had a rsync transfer going on via cygwin. I forgot something, so I issued a STOP SLAVE command to MySQL. This caused the entire server to literally hang. No reply on ping, nothing. After about 15 minutes in this state, the box was manually hard rebooted.

This raises the question in my mind if I can trust the server. STOP SLAVE isn't an intensive call at all. It's beyond me why that would cause MySQL to crash, yet alone the whole operating system. So now I'm wondering if it's a hardware problem. We just got brand new Ram (32gb) installed in the server, but they never ran memtest on it. Since I don't have physical access to the server (in a different country), they won't run memtest until Monday morning. I want to do as much testing over the weekend as I possibly can.

I had a similar issue in Linux a few years ago which was caused by a faulty bios, where under high I/O loads the box would just freeze. What I did then to reproduce it was have a few python scripts generate a number of large (10gb+) files, and then randomly seek to different positions among those files. This caused the machine to halt within minutes.

So that got me thinking, why not do a similar thing. So I wrote a python program to read and write to a series of files (running in 4 processes) to hopefully saturate the disks. Then I wrote another one to just try eating as much ram as possible (it's at 32gb now and climbing) while randomly reading and writing to positions in its list. It's been cranking for about an hour now, and still solid (the swapping is slowing things down, but it's still stable).

So I come here to ask, are there any user-land ways of stress testing 2k8 that aren't really application dependent? Once MySQL catches up, I'll write a script to randomly query that to increase the I/O and memory workout. But I'm more looking to test the machine and OS more than the application... But until that point, I want to punish this machine for the halt.

Thanks

Here's the background on our situation...

Right now, we are setup as three distinct companies with three complete Active Directory and Exchange systems. The three offices (One in the US, two in Europe) are connected via a three way VPN setup (so each office has secure communication to the other two). There is a two-way trust relationship setup in Active Directory for each setup. All systems are running Server 2003 and Exchange 2003.

There are about 160 mailboxes between the companies and 80 users (the additional mailboxes are either for IT subsystems, forwarding accounts or other uses).

The companies are officially merging together (instead of just having a trust relationship). So we're looking into a combined solution (based on a new name) where each office will be on the same systems (Exchange and Active Directory) as well as consolidating our IT infrastructure (there's a lot of duplication).

They hired an external company to come in and audit our IT infrastructure. They have made an official recommendation to outsource the IT infrastructure (and guess what, they want to provide the service).

I've been tasked with figuring out what to do. I've thought about it quite a bit, and I've come up with two options. The basic difference is where Exchange is hosted (internally our outsourced). Since outsourced is easy to fathom, I'll just detail the internal setup.

Since high availability is required, we want some geographic redundancy built in. So, what I've come up with is as follows (I'll call the offices Site1, Site2 and Site3):

Site1:

• FSMO Active Directory Role
• Exchange Mailbox Role - Primary
• Exchange Client Access, Hub Transport Server Roles
• DFS File Share Role (for shared drives)

Site2:

• Active Directory Role - Replicated from Site1
• Exchange Mailbox Role - Secondary, replicated using CCR Replication
• Exchange Client Access, Hub Transport Server Roles
• DFS File Share Role

Site3:

• Active Directory Role - Replicated from Site1
• Exchange Client Access, Hub Transport Server Roles
• File Share Witness (for failover)
• DFS File Share Role

So basically the cluster should be able to survive a single site failure without bringing down any of the other sites (or any of the systems). In the event of a double site failure, Exchange would stop completely.

So, my concerns are as follows:

1. Is this a reasonable setup? Or am I over complicating things?
2. The number of servers required (3 at each site since CCR Mailbox roles must be the only role installed).
3. Will it even work as summized (where it will automatically fail-over to the available node should a site or server go down)?
4. Since each office would specify a local Client Access server for its users, that server becomes a single point of failure for all local requests (But this is solvable by a manual DNS change)
5. Do all of these servers need to be on the same IP subnet for this to work? Or can I get away with using a hiearchial DNS for it (clientaccess.site1.foo.com, etc)?
6. This will let me set each office as an MX record (since there is a hub transport server in each office to connect to the internet) so if one office goes down we still should be able to receive email in the others, correct?
7. Maintainability. I have a fear that this setup will be too complicated to maintain in the long run (adding offices, removing offices, upgrading servers (both OS and hardware), etc). Is that a justified fear?

Now, there's also the question on whether to go with server 2003 or 2008... If we go the internal Exchange route, I think I can convince the powers to upgrade to 2008 (in fact we would need to upgrade to use Exchange 2010)... But is it really necessary or is that just one of my "wants" sneaking in to the plans (rather than a justifiable upgrade)...

Now, part of me just wants to go with outsourced Exchange since it'll alleviate some of these issues (or most of them). However after looking at the costs, the break-even point is about 1 year, so after that outsourcing will be considerably more expensive. Couple that with the fact that some features we depend upon are not possible outsourced -- at least with the companies we looked at-- (such as Shared Mailboxes, Active Directory coupling including SSO, centralized management, data security, etc). So I'm really torn as to where to go with this...

This is the first project of this scale that I'm attempting, so any help would be greatly appreciated...

Thanks in advance (and sorry for the book)...

I'm in the process of setting up an Openfire XMPP server for intra-company communication. I have it authenticating against our local DC fine. Our company has two different offices (one in the US, and another in Europe) each with its own domain. Our two systems are fully trusted and connected via a VPN.

What I would like to do is setup Openfire to allow users from either domain to log in to the single primary server. I'd like to keep the group chat feature, so installing a server on each domain isn't really a great option (but it's open if I can't get this working).

I can see all the trusted users in Openfire (It's detecting them for the contact lists), but I can't get them authenticating (They get a Not Authorized. Please Try Again message).

Any thoughts on how to proceed?

I've been having some issues over the past few days with our WSUS server eating 100% CPU usage. We don't have many clients (maybe 30 or so at most), so that's not the issue (Not to mention that this server had more in the past).

So, in my debugging, I've completely re-installed both WSUS and Windows Internal Database (SQL Server 2005) several times (Nuking the db and download files each time), and keep having the same issue. sqlservr.exe uses around 85% of the available CPU, and wsusservice.exe uses the other 15%. Keep in mind that basically nothing is happening on the server at this point (no clients connected, etc).

I opened up task manager, and I see that between the two processes, they are using around 275 mb of ram (on a 1024 mb system). However, the page faults just keep racking up (sitting at around 7k to 10k per second). So, in an attempt to see if that was the problem, I disabled the page file in windows and restarted. I'm still seeing the high rate of page faults. And the performance tab of task manager still shows about 600mb worth of PF usage... What I don't understand, is why is it still tracking PF usage if there is no paging file...? (The drives are thrashing, so something is going on with the disk, and I can't see very much disk I/O in task manager. And the VMWare manager is showing a moderate write rate since I disabled the PF, about 1.5 Mbps. It's also showing a falloff in the read rate, 9 Kbps. The drives are capable of burst rates much higher than that, so I can only assume it's due to seeks why it's so low (random writes)).

The Commit Charge is as follows:

1. Total: 634408
2. Limit: 960004
3. Peak: 657512

The Physical Memory is as follows:

1. Total: 1048012
2. Available: 21864
3. Cache: 402168

Oh, and it's running Server 2k3 SP1 on a VMware ESXi 3.5 server (Quad Xeon, 8gb ram and 6 x u320 SCSI HDD).

I'm lost at this point as to how to continue with this...

Actually, in the time I've written this, the sqlservr.exe process's memory usage has jumped from around 100mb to 330mb. It's doing a synchronization now, so I can understand the increase, but still it never used to eat up 100% CPU for something simple like a full sync...

We have a few users who are sharing a rather large excel file (70 megs) via a network drive. If they save the file after each tiny modification, then there are no problems. But if they try to do a bunch (as little as 10), Excel hangs when they try to save it. Plus, the file becomes locked, so the other needs to rename the file and re-save his work. I've tried to search for this issue, but have had little to no luck.

The network drive is on a local network server (Win Server 2k3, gigabit networking) with plenty of free space, so that shouldn't be the issue. I'm kind of at a loss as to where to go from here. Is this a bug with Excel? Is it just something we're going to have to live with? Or is there something else that I/We are overlooking?

Thanks in advance...

I've got an interesting one for you. I've been going over my server monitoring and notification systems (Nagios based), and realized that if our internet connection goes down, there's no way for it to notify me. I already have a modem listening (Via CentOS 5) on a spare POTS line so that I can dial-in in case our internet goes down. I was wondering if I could come up with a script (Shell, Python, etc) that can dial out and play a recorded message (wave file I'm guessing) when it's picked up. I know Windows supports voice calls over a voice modem, I was wondering if a solution existed for Linux...

I know asterisk can probably do it, but isn't that overkill (A full blown VOIP system just for a notification mechanism that will hopefully never be used)? And wouldn't it interfere with the modem's primary function as a backup network interface (PPP spawned via mgetty)?

I've done some searching, and haven't really come up with much. I know how to dial out from the command line, but only as a modem (not as voice). Worst case, I could set it up to dial out as a modem, and then just realize that if I get a call with modem sounds from that number that it's the notification... Any insight would be appreciated...

I have a forward setup on one of our public facing email addresses to a few users outside of our system (to an external address). Seeing as it's public (printed on a website in plain text), we get TONS of spam on that account. We're being marked as a SPAM source due to the forwarding by the external system. We have pretty aggressive IMF (Intelligent Message Filtering) setup, and it works great for internal messages. The problem we're facing is that Exchange is forwarding ALL the email received on that account, and not just the messages that pass the IMF (and don't get pushed into Junk).

As far as I can tell, there's no way to adjust the IMF settings for that one account. And as far as I can tell, there's no way to setup folder forwarding (inbox -> external address) from Exchange. Is there something I'm missing or is there any way around these issues that I may be missing?

Thanks...