Has anyone noticed that Verizon’s choices of phones are lacking? I mean if you want LTE, you currently have 2 choices: The HTC Thunderbolt or the DROID Charge. Both of these phones are single processor 1 GHZ phones. AT&T has the Atrix and T-Mobile has the G2X which both use the Tegra 2 (Dual core 1GHZ ARM processor). If you are thinking of iPhone, you need to remember that with Verizon you cant use 3G data and voice at the same time. It isn’t that the technology isn’t there, it is just that Verizon chooses not to use it (Verizon could be using SVDO instead of EVDO to fix that problem).

I guess what I’m trying to say is that in my opinion there isn’t a good phone on the Verizon network. I wish Verizon understood that a good network with crappy phones is a crappy option. So when is this going to change?

I’ve been doing a lot of thinking lately on RAID. We are looking into upgrading some servers. I noticed that Microsoft tends to recommend RAID 10 over RAID 5. I spent a lot of time thinking about why they’d recommend RAID 10 over RAID 5. I do understand that when a drive fails with RAID 5, during the rebuild performance is effected much greater than it would be with RAID 10. I also understand that RAID 5 is also more vulnerable than RAID 10 as well (RAID 5 can only withstand a single drive failure. RAID 10 can lose up to half of the drives as long as the original and mirrored drive don’t die).

RAID 5 had its own advantages though. Redundancy doesn’t cost you half of the total disk space. Sequential Reads and Sequential writes can be faster with a decent RAID controller. It has to calculate the parity, but the only penalty is writing the parity to disk (1 penalty for writing each stripe. Where as with RAID 10, you have to write every block twice). Random Reads shouldn’t be effected too much because they could end up reading from any of the disks in the array.

The problem area for RAID 5 is random writes. You can see it in the benchmarks listed here.  If we pick a random part of the disk and decide to write a 4KB block, with RAID 10, we just need to write that same block to the same area on the mirrored disk. Thus we incur a penalty of 2 (We have to write the block twice) when writing a random block for RAID 10. On RAID 5 however, we cannot just write a random block, but must also recalculate the parity block and rewrite it to the disk. In order to calculate parity, we have to read what the block being overwritten is, read what the parity block is. Then compare the old block to the new rewritten block. For each bit that is changing, we must switch the parity. Anyway, with a RAID 5 array of 4 disks and larger, you incur at least 2 reads and 2 writes (Or 4 I/O operations compared to RAID 10′s 2 I/O operations).

I’ve been looking at server upgrade pricing and features of Windows Server lately. I was digging through Microsoft website when I ran across the Edition Comparison by Technical Specifications. Apparently Windows Server 2008 R2 standard is limited to 32 GB of RAM! Windows 7 Pro x64 is limited to 192GB of RAM! Given Server can do a lot of things that Windows 7 can’t, but 32 GB of RAM is a low limit for a server grade product.

32 GB of RAM is quite limited since they allow 4 processor sockets. That is a mere 8 GB of RAM per processor. Intel server processors typically use triple channel memory, while AMD’s newest processors use quad channel memory. This means if you have 2 Intel processors and want to have the same amount of memory for both processors, while optimizing memory (Means all 3 channels are filled with the same amount of memory per processor), then you are limited to six 4 GB and six 1 GB memory sticks (30 GB total). With AMD, we are limited to eight 4GB sticks. It is easy to see that buying 8 and 16 GB sticks are out of the question in this situation with Server Standard.

So what am I going to do about it? —- I’m not going to buy Standard. The Enterprise version of Windows Server is limited to 2 TB of RAM. It also allows for you to run 4 VMs of Windows Server on the system that it is physically installed (vs 1 on Standard). It also has fail-over clustering which is nice if you are running Exchange (It means you can add the server into a Database Availability Group). Enterprise is quite a bit more expensive though ($3999 vs $1209), however Enterprise comes with 15 more Windows CALs than Standard (valued at about $600), so Enterprise will only cost you about $2200 more than Standard.

1. A Backup Schedule. This implies you have a backup plan. RAID isn’t a backup. Backups are useful when you have a catastrophic hardware failure, accidental deletion/overwriting of files, or a disaster situation. One other question you should ask is how much data loss is acceptable. For instance: If your backup media only goes off-site once a week, then you have the potential of losing up to a week of data (If there is a disaster right before the media goes off-site). If data goes off-site every day, then you may have up to a day of data loss.

   Capacity comes into play here too. How much data do you need to backup? How much money do you have to spend on backups? How much data loss is acceptable? Can you save money by doing a full backup every once in a while and doing incremental backups other days/weeks (Weeks would probably be risky)?

2. A Disaster Recovery Plan. This goes along with having a backup. Your data may be backed up, but are you sure that you can restore all data and services from the backup media? If you’ve not tried restoring from backup media, I strongly encourage that you do a trail run. I did so with my business and found issues with restoring Active Directory and Microsoft Exchange. I was able to document and address these issues now —- Instead of dealing with them in a disaster situation that is much more stressful. I didn’t need a bunch of servers to test disaster recovery. I did so with a couple of decent desktop computers and VMware Server (It is free!)
3. UPSes and the ability to shut down servers automatically during a power loss. Simply having a UPS isn’t enough. If the servers are never made aware that their is a power outage, then they’ll just continue to run until they exhaust the UPS battery. Unexpected server shutdowns can cause data loss. The only thing just a UPS will do is preventing data loss from short power outages. Most server grade UPSes have the option to purchase an add-in network card. UPS vendors will typically provide software which allows servers to monitor UPSes and shut down in the event of a power loss. We use APC. The Smart-UPS series UPSes have a card slot on the back of them. We’ve purchased an AP9617 (I recommend getting the AP9618 though) which has an Ethernet port. This allows a single UPS to communicate to several servers.

   Power is another issue that shouldn’t be overlooked. Plugging 2 1500VA UPSes into a single 20 amp circuit can kick a breaker and cause a power issue. Be sure you plan to have enough power available so your servers don’t kick breakers. Plan extra capacity in your design. After a power failure, UPSes have to charge AND your servers have to run at the same time.

4. A plan to shut down servers in the event of an environmental problem. I’m speaking mostly about heat related issues, but it is a good idea to ensure that you don’t have a water sprinkler in your server room (Or in the area that your server is located). In the event that there is a sprinkler in the location, you may make management aware of the dangers. If your business is building/renovating a location, I’d recommend getting a waterless fire suppression system.

   Anyway, back to the topic, heat. I have a friend that works for a large corporation. They rent space at a large data center. Anyway, at one point they lost power on a single phase of their three phase power. This wasn’t enough to disable power to the servers, but it did disable power to their AC. Temperatures quickly climbed to 180° F. This did kill some of their servers, but it also wounded many more. These server exhibited odd behavior and would die off slowly in the coming weeks. If the AC stops functioning in your server room (Or in the room that your server is located), what do you have in place. In the previous point, I recommended the AP9618 add-in card to APC UPSes because it has environmental monitoring capabilities and can shut down your servers when it gets too hot in your server room.