(Not logged on) | Register | Log On

You can subscribe to this discussion group using an RSS feed reader. The Joel on Software Discussion Group

A place to discuss Joel on Software

This community works best when people use their real names. Please register for a free account.

Other Groups:
Joel on Software
Business of Software
Design of Software (CLOSED)
.NET Questions (CLOSED)
TechInterview.org
CityDesk
FogBugz
Fog Creek Copilot


The Old Forum


Your hosts:
Albert D. Kallal
Li-Fan Chen
Stephen Jones

What about blades?

When creating your new Server Architecture have you considered using blades?

What are the pros and cons of using blades in an architecture similar to what Joel is proposing?
Demis Bellot Send private email
Sunday, February 06, 2005
 
 
Does anybody actually use blades? I thought they were mostly hype, and that most places just used pizza boxes for that kind of purpose (cheap-ass webservers).
bof.
Sunday, February 06, 2005
 
 
With a blade architecture isn't the chassis a single point of failure?

Philo
Philo Send private email
Sunday, February 06, 2005
 
 
Aren't they still clustered so reliable?
son of parnas
Sunday, February 06, 2005
 
 
There's ALWAYS a single point of failure (I know that's not 100% true, but most of the time, it is). If it's not the circuit, if it's not the power supply, maybe it's the motherboard. What about your network card? Do you have dual, redundant network cards? Do you have two separate networks, so that failures of switches aren't fatal?

I know not all of these are really issues, but I doubt you're going to be paying for the system with zero single points of failure. The question is not whether a single point exists, but what the likelyhood of that failure is.
Anon to Protect the Guilty
Sunday, February 06, 2005
 
 
In our case we decided against blades because of

(1) indeed the chassis is a single point of failure
(2) space in the colo is not SO expensive that I'll pay a premium to save a couple of U's
(3) i really like having 6 hot swap HD slots in each server; this has been very useful on our office servers
Joel Spolsky Send private email
Sunday, February 06, 2005
 
 
Anon, thats not always true.  Most good setups don't have a single point of failure, usually multiple web servers with different interfaces, and a master-slave DB is pretty good.  The problem Philo brought up, which is a good ones, is the fact that now your 'application server cluster' now has a single point of failure.

I've looked at blades before, and they just seem really expensive, and I don't really see what the advantage is besides space.  I suppose if form factor is at a premium, they would be worth it, but a single blade is more then the equivelant 1U, and you have to spend a lot of money on the chassis.  I didn't look too much into this, so if I'm wrong, I'd love to hear positive experiences with them.
Vince
Sunday, February 06, 2005
 
 
"I'd love to hear positive experiences with them."

Blades can put a lot of compute power in a small area.  From an Oct 2002 article:

"BioLink will deploy 456 800i single-processor Linux blade servers from RLX Technologies Inc. to voter-registration sites in Nigeria's 37 states. The blades will process data from scanned voter-registration cards, which will include voters' thumbprints. BioLink's software will run on the blades, checking for fraud or duplication."

http://www.informationweek.com/story/IWK20021004S0002
Mark Jerde Send private email
Monday, February 07, 2005
 
 
My company uses blades.  Our blades do have redundant power supplies, nics, etc; but there is a single point of failure in the chassis somewhere (the backplane??).
One of the reasons we moved to blades is reduced hardware support costs.  We have 1000+ servers.  With blades, there are fewer total power supplies, network cards, etc.  I'm sure there is a break even point where it makes sense to start using something like blades, but I'm not sure how high (or low) that number is.  IBM or HP probaly have case studies on this.
Also, I'm not sure the disk drive argument is completely valid - if you have a disk array attached to the blades, you can still have a large number of hot swappable drives.  I think the boot drive must be local, but I believe it can be mirrored also.
bw
Monday, February 07, 2005
 
 
Blades can reduce costs by allowing several servers to share some redundant equipment.  Yes, the chassis does become a single point of failure all 6 servers, but if the organization is large enough to need several sets of blade servers, then that single point is eliminated also.  For smaller applications they just don't make sense.
Joel Coehoorn
Monday, February 07, 2005
 
 
For smaller organizations blades are probably only useful for saving on the energy bill or some closet space. Either send that money to your energy company or HP--your choice. But for micro ISVs where their colocation and inhouse IT networks are a big piece of investment--you still win the biggest with standard name-brand PC form factors--17" and 19" towers. Pick the sweet spot in Dell's business desktop line, add a ton of brand name memory/HDs (from your local PC outlet) and then pray. 1-5 of these power desktop should handle apache/tomcat/postfix/proftpd/bind/samba/rsync-over-ipsec or iis/aspx/exchange/windows dns/bulletproof/backupexec without breaking a sweat for a workgroup of 5-50 people*.

* it might be a problem stuffing everything into the same box--for security.
Li-fan Chen Send private email
Tuesday, February 08, 2005
 
 

This topic is archived. No further replies will be accepted.

Other recent topics Other recent topics
 
Powered by FogBugz